# We test ChatGPT Images 2.0: a comparison with Nano Banana Pro and new uses for government, politics and public administration

Author: Fernando Nieto Lobato
Original publication: 2026-04-29
Spanish original: https://estrategiabyaleph.substack.com/p/estrategia-135-de-nano-banana-pro
English URL: https://elcontemplador.github.io/estrategia-english/essays/135/
Status: Published translation

This is a translation of the original Spanish essay published on 29 April 2026. Its claims, examples and forecasts retain that historical context.

English publication: 2026-09-29

*Archive note: this article reports experiments published on 29 April 2026. Model availability, rankings and expectations retain that date. The generated posters, statistics and procedural advice below are historical demonstrations, not verified official information. English transcriptions reproduce their visible content, including anomalies.*



Just a few days ago, on 21 April, [OpenAI introduced ChatGPT Images 2.0](https://openai.com/es-419/index/introducing-chatgpt-images-2-0/), its new image generation and editing model **integrated into ChatGPT**. The tool can already be used within the conversation itself, including on free accounts, although with usage limits and some advanced functions reserved for paid plans. As usually happens with these launches, it is wise to keep a little distance from the initial enthusiasm, but **our first tests point to a clear improvement**: greater visual coherence, a better understanding of complex instructions, more orderly composition and, above all, an increasingly reliable ability to integrate legible text into images, which makes a real difference for many uses. The first comparisons by [Artificial Analysis](https://artificialanalysis.ai/image/models?submodality=text-to-image) and [Arena AI](https://arena.ai/leaderboard/text-to-image) also place it at the top of image generation, ahead of GPT Image 1.5 (the previous model), Nano Banana 2 and Nano Banana Pro (Google’s models), **by a considerable margin**.



[![Artificial Analysis Image Arena quality Elo chart, with all fifteen model scores transcribed below.](https://elcontemplador.github.io/estrategia-english/assets/images/be11b23a-66b3-486c-962c-b1ec9ca0f5dd_2048x732.png)](https://substackcdn.com/image/fetch/$s_!1wuy!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbe11b23a-66b3-486c-962c-b1ec9ca0f5dd_2048x732.png)



*Accessible transcription of the original English chart: “Image Arena Quality ELO”. Elo scores from blind preference votes in Artificial Analysis’s Image Arena.*



| Model, as labelled | Score |

| --- | ---: |

| GPT Image 2 (high) | 1332 |

| GPT Image 1.5 (high) | 1271 |

| Nano Banana 2 (Gemini 3.1 Flash Image Preview) | 1262 |

| Nano Banana Pro (Gemini 3 Pro Image) | 1217 |

| Seedream 4.0 | 1204 |

| FLUX.2 (max) | 1203 |

| MAI-Image-2 | 1196 |

| FLUX.2 (pro) | 1188 |

| grok-imagine-image | 1186 |

| FLUX.2 (flex) | 1182 |

| ImagineArt 2.0 | 1179 |

| Imagen 4 Ultra | 1176 |

| FLUX.2 [dev] Turbo | 1168 |

| Seedream 4.5 | 1167 |

| Nano Banana (Gemini 2.5 Flash Image) | 1162 |



[![Arena AI average text-to-image win-rate ranking, with fifteen values transcribed below.](https://elcontemplador.github.io/estrategia-english/assets/images/e4ecf104-bd52-4659-b182-1ccf3c750438_1080x1081.png)](https://substackcdn.com/image/fetch/$s_!NZjQ!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe4ecf104-bd52-4659-b182-1ccf3c750438_1080x1081.png)



*Accessible transcription of the original English chart: “Image Arena — Model Win Rate: Average Text-to-Image”. Source: Arena AI leaderboard. Average win rate, ties excluded; the reference line is 50%.*



| Rank | Model, as labelled | Win rate |

| ---: | --- | ---: |

| 1 | GPT Image 2 (Medium) | 93% |

| 2 | Nano Banana 2 | 67% |

| 3 | Nano Banana Pro | 64% |

| 4 | Nano Banana Pro (2k) | 62% |

| 5 | GPT Image 1.5 (High) | 62% |

| 6 | Reve V1.5 | 52% |

| 7 | MAI Image 2 | 50% |

| 8 | Grok Imagine Image | 47% |

| 9 | Flux-2 Max | 45% |

| 10 | Flux-2 Dev | 45% |

| 11 | Hunyuan Image 3.0 | 44% |

| 12 | Grok Imagine Image Pro | 44% |

| 13 | Flux-2 Flex | 44% |

| 14 | Nano-Banana | 43% |

| 15 | Flux-2 Pro | 43% |



That is why **[in this issue we wanted to run a direct test: take the vast majority of the prompts we used in estrategIA #115 — “Ten strategic uses of Nano Banana Pro for communicating public policies, campaigns and services”, published on 10 December 2025 — and run them again with ChatGPT Images 2.0](https://estrategiabyaleph.substack.com/p/estrategia-115-diez-usos-estrategicos?utm_source=publication-search)**. The aim is not to decide which model “draws prettier pictures”, but to see whether this latest leap makes more robust uses possible in political communication, government and public administration. We encourage you to open the earlier issue and compare the results from the two models, bearing in mind that these are the first results produced by each prompt: we neither edited them nor made several attempts in either case. **After that comparison, we also bring you several new cases that we think are particularly interesting.**



## Some uses from [the December issue](https://open.substack.com/pub/estrategiabyaleph/p/estrategia-115-diez-usos-estrategicos?utm_campaign=post-expanded-share&utm_medium=web), revisited for comparison



**[All the prompts are in the original article](https://open.substack.com/pub/estrategiabyaleph/p/estrategia-115-diez-usos-estrategicos?utm_campaign=post-expanded-share&utm_medium=web)**; we will not repeat them and make this newsletter even longer, but they are available there for you to run your own tests and designs. Of the ten uses we presented in December, we have chosen **five examples** to show below, again for reasons of space. However, **we have tested almost all of them, and the new model handles them without difficulty and, in our view, with higher quality; its handling of text stands out in particular**.



### Producing emergency posters quickly



[![Generated orange weather-alert poster showing a snowy city and advice to avoid travel.](https://elcontemplador.github.io/estrategia-english/assets/images/e46c142b-f29b-483b-9f95-4043d0ded134_1122x1402.png)](https://substackcdn.com/image/fetch/$s_!a92U!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe46c142b-f29b-483b-9f95-4043d0ded134_1122x1402.png)



*English transcription of the generated poster:* “Weather alert — Heavy snowfall. Avoid travel. Heavy snow; reduced visibility; snow accumulation. Risk on roads and in urban areas. Get information through official channels. Your safety is our priority.” The illustration shows a snow-covered urban neighbourhood, vehicles and a snowplough.



### Communicating the benefits of a new policy to citizens



[![Generated Salamanca social-housing policy poster with objectives, figures and implementation dates.](https://elcontemplador.github.io/estrategia-english/assets/images/0e4cc27c-6556-4d4d-9160-b01e5c30306b_941x1672.png)](https://substackcdn.com/image/fetch/$s_!Qr7c!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0e4cc27c-6556-4d4d-9160-b01e5c30306b_941x1672.png)



*English transcription of the generated example; its council branding and figures are reproduced as model output.*



**Salamanca City Council — Public policy: building new homes for social rent in Salamanca.** More affordable homes for a fairer, more inclusive and cohesive city.



| Panel | Content |

| --- | --- |

| Main objective | Increase Salamanca’s public stock of social-rent housing to give those who need it most access to decent, affordable, high-quality homes. |

| Who is affected? | People and families with limited incomes who are registered as residents or work in Salamanca, with particular attention to young people, older people, single-parent families and vulnerable groups. |

| Expected benefits | Better access to affordable housing; a lower financial burden on households; retaining the population and revitalising neighbourhoods and districts; greater social cohesion and equal opportunities. |

| Three key figures | 300 new social-rent homes; €9.6 million estimated total investment; more than 900 people and families benefiting. |



**Implementation timetable:** 2024 — planning and design: needs assessments, selection of land and projects. 2024–2025 — processing and tendering: project approval and award of construction contracts. 2025–2026 — start of works: construction of the new homes begins. 2026–2027 — construction: works and quality controls. 2027–2028 — handover and letting: allocation of homes and bringing them into use.



“Committed to the right to housing and the future of Salamanca. Homes for people. A city for everyone.” The displayed contact details are www.aytosalamanca.es/vivienda, Municipal Housing Office, C/ Pozo Amarillo, 2 · 37001 Salamanca.



### Creating an infographic about an administrative procedure



[![Generated DNI renewal infographic with model-produced requirements, warnings and example identity card.](https://elcontemplador.github.io/estrategia-english/assets/images/1f66d8f4-6f48-42b2-a6e6-931f85504b34_1024x1536.png)](https://substackcdn.com/image/fetch/$s_!W2YC!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1f66d8f4-6f48-42b2-a6e6-931f85504b34_1024x1536.png)



*English transcription of the generated example. The requirements, fees, validity periods and contact details below are claims within the model’s demonstration, not verified instructions for renewing an identity card.*



**Renewing your DNI:** keep your identity up to date. Your DNI identifies you, protects you and opens doors for you. Renewing it is quick and simple. Do it before it expires.



The sample Spanish national identity card reads: surnames García Pérez; first name María; date of birth 01/01/1990; DNI number 12345678A; valid until 01/01/2030; signature.



| Step | Instructions in the image |

| --- | --- |

| 1. Make an appointment | Request an appointment on the official website or through channels provided by the administration. Official website: www.dnielectronico.es. Telephone: 060. |

| 2. Go to the office | Attend the issuing office in person at the date and time of your appointment, with all the documents. National Police offices authorised to issue the card. |

| 3. Required documents | Your previous DNI; one recent colour photograph with a light background, 32 × 26 mm; evidence of any changed details, such as name, surnames or address. Check specific requirements on the official website. |

| 4. Fees | Pay the renewal fee. The image says the amount may vary by autonomous community or place of payment. Cash or card, depending on the office. |

| 5. Validity | The new DNI will have the validity period established for your age. Always check the expiry date. |



**Common warnings:** do not arrive without a photograph — bring one recent colour photograph with a light background and without dark glasses; the image says photographs are not taken at the offices. Renew before expiry; if the card has expired, additional requirements and fees might apply. Age exceptions listed: under five, valid for two years; ages five to thirty, five years; over thirty, ten years.



**Useful tips:** check the expiry date in advance to avoid problems; verify your appointment’s date, time and office and arrive a few minutes early; check that your personal details are correct before completing the procedure; find out the cost and accepted payment methods and bring the exact amount; if you have moved, bring updated proof of address; if in doubt, consult the official website or call citizen-service channels.



“Your DNI is your identity. Renewing it on time is your responsibility and your security.” More information: www.dnielectronico.es. “This procedure must be completed in person. Nobody can do it for you. Protect your identity and your data.”



### Making storyboards for videos



[![Generated nine-scene political advertisement storyboard with timings and text.](https://elcontemplador.github.io/estrategia-english/assets/images/84fbd58c-3559-46ce-8f25-3fd112341f56_1149x1369.png)](https://substackcdn.com/image/fetch/$s_!Jm0d!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F84fbd58c-3559-46ce-8f25-3fd112341f56_1149x1369.png)



*English transcription of the generated storyboard: “Political advertisement — Together, we build tomorrow”. Hopeful tone; estimated duration 60 seconds; format 16:9. The individual durations shown below total 64 seconds, an inconsistency preserved from the image.*



| Scene and shot | Visual narrative and duration | Audio, on-screen text and emotion |

| --- | --- | --- |

| 1. Dawn — wide shot | The sun rises over the city. A new morning begins. 6 seconds. | Soft, inspiring music. “Every day is a new opportunity.” Hope, renewal. |

| 2. Our people — medium shot | Different people in everyday life: working, studying, caring, creating. 8 seconds. | Warm voice-over. “Different stories, one shared dream.” Identification, connection. |

| 3. Listening — medium shot | The candidate listens to citizens in different neighbourhoods or communities. 8 seconds. | Voice-over. “We listen to understand.” Trust, respect. |

| 4. Proposals in action — wide shot/short shots | Real projects or depictions: construction, education, health, environment, innovation. 10 seconds. | Energetic voice-over. “Clear proposals. Real action. Results for everyone.” Trust, optimism. |

| 5. United, we move forward — wide shot | Diverse people walking together, collaborating in their community. 8 seconds. | Music builds. “Together, we are stronger.” Unity, belonging. |

| 6. Looking to the future — wide shot | Children look towards the future in a clean, modern setting: park, school, sustainable city. 6 seconds. | Voice-over. “Let us build the future they deserve.” Hope, inspiration. |

| 7. The commitment — close-up | The candidate looks at the camera with determination and sincerity. 6 seconds. | Candidate’s voice. “My commitment is to you.” Credibility, connection. |

| 8. Together, we succeed — wide shot | The community celebrates an achievement: an opening, a cultural event, a lively square. 6 seconds. | Music at its peak. “Together, we succeed.” Joy, pride. |

| 9. A memorable ending — graphic/wide shot | Logo, candidate’s name and final message against a city at sunset. 6 seconds. | Closing voice-over. “A better tomorrow is in our hands. Let us keep moving forward.” Hope, motivation. |



The closing graphic also says “Candidate’s name — Together for our community”. Central message: “We listen. We work. We deliver. For you, for our community, for the future.” Overall tone: positive, inclusive and hopeful.



### Generating explanatory infographics based on data



[![Generated Spanish unemployment infographic with annual values, timeline and international comparison.](https://elcontemplador.github.io/estrategia-english/assets/images/472a09ca-8c75-4665-b511-e49ce0cda558_1024x1536.png)](https://substackcdn.com/image/fetch/$s_!wiMa!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F472a09ca-8c75-4665-b511-e49ce0cda558_1024x1536.png)



*English transcription of the generated infographic. All values, dates and source labels are reproduced as displayed; they have not been independently validated. Some timeline headings name a different year from the year of the event they describe.*



**Unemployment in Spain, 2010–2025:** unemployment rate as a percentage of the labour force, and key political and economic milestones. Displayed average for 2010–2025: **16.1%**. From the 2013 peak of 26.1%, unemployment has fallen **15.9 percentage points, or 62%**.



| Year | Unemployment rate |

| --- | ---: |

| 2010 | 19.9% |

| 2011 | 21.4% |

| 2012 | 24.8% |

| 2013 | 26.1% |

| 2014 | 24.4% |

| 2015 | 22.1% |

| 2016 | 19.6% |

| 2017 | 17.2% |

| 2018 | 15.3% |

| 2019 | 14.1% |

| 2020 | 15.5% |

| 2021 | 14.8% |

| 2022 | 13.0% |

| 2023 | 11.6% |

| 2024 | 10.6% |

| 2025* | 10.2% |



*2025 data are labelled as covering the first quarter.*



| Timeline year shown | Milestone and explanation displayed |

| --- | --- |

| 2010 | 2010 labour reform: emphasis on internal flexibility and employment stability. |

| 2011 | Sovereign debt crisis: increased uncertainty and fiscal tightening. |

| 2012 | Economic recession: sharp fall in activity and job losses. |

| 2013 | Historic unemployment peak: 26.1% of the labour force. |

| 2014 | 2012 labour reform: begins to produce results in hiring. |

| 2015 | Signs of recovery: confidence improves and more jobs are created. |

| 2016 | Political stability: government formed after a repeat election. |

| 2017 | Toledo Pact: recommendations to strengthen pension-system sustainability. |

| 2018 | Solid growth: employment grows at a good pace and unemployment continues to fall. |

| 2019 | Social dialogue: agreements to improve employment and wage quality. |

| 2020 | COVID-19 pandemic: severe impact on activity; unemployment temporarily rises to 15.5%. |

| 2021 | Recovery Plan: Next Generation EU funds to boost recovery and employment. |

| 2022 | 2021 labour reform: comes into force, with greater employment stability and less temporary work. |

| 2024 | Slowdown with resilience: the labour market holds up despite weaker growth. |

| 2025 | Best level since 2008: unemployment reaches 10.2%, the lowest in the series since 2008. |



The phases shown are: 2010–2013, crisis and job losses; 2014–2016, start of recovery; 2017–2019, expansion and job creation; 2020–2021, pandemic shock and response; 2022–2025, consolidation and labour-market improvement.



**Key points displayed:** unemployment peaked in 2013 at 26.1%; it has declined continuously since then, with a slight rise in 2020 because of the pandemic; labour reforms, social dialogue and European funds have been crucial to improvement; the 2025 rate of 10.2% is the lowest since the series began in 2008. **Figures:** −15.9 percentage points since 2013; −62% relative to the 2013 peak; +3.7 million jobs created since 2013, labelled “EPA, Q1 2025”.



**International comparison, 2024**, unemployment as a share of the labour force: Spain 10.6%; euro area 6.4%; Germany 3.1%; France 7.4%; Italy 7.7%. The image labels the comparison’s source “Eurostat (EPA), 2024 data” and the overall source “INE (Labour Force Survey)”.



## Five examples of new uses with ChatGPT Images 2.0



Since the model, as the benchmarks clearly show, represents a significant leap forward in image generation — we do not think Google will take long to respond with another iteration of its models, probably at its major developer conference, Google I/O, in mid-May — we bring you **new possible uses in politics, public administration and government**, together with prompts we co-generated with GPT-5.5.



### An infographic for a crisis plan



*Prompt, translated from the Spanish used in the experiment:*



> “Crisis room: the first six hours of a viral falsehood about a public institution.” The image should represent an institutional coordination centre, not a chaotic scene. Combine a visual scene of a crisis room with a clear operational panel. Include a six-hour timeline: Hour 0: detection of viral content. Hour 1: internal verification. Hour 2: consultation with official sources. Hour 3: response decision. Hour 4: publication of a clarification. Hour 5: impact monitoring. Hour 6: report and learning. Add three side columns: responsible people; critical decisions; mistakes to avoid. Style: serious, institutional, realistic and modern, with the appearance of an operational protocol for a government office. Perfectly legible Spanish text. Do not use the names of parties or real people.



[![Generated crisis-room plan with responsible teams, a six-hour timeline and decisions.](https://elcontemplador.github.io/estrategia-english/assets/images/00c0157f-e19b-48b4-bae1-1f6b87972ad3_1672x941.png)](https://substackcdn.com/image/fetch/$s_!bSvB!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F00c0157f-e19b-48b4-bae1-1f6b87972ad3_1672x941.png)



*English transcription of the generated crisis plan: “Crisis room: the first six hours of a viral falsehood about a public institution”. Coordination · verification · transparency · responsibility.*



| Responsible team | Role |

| --- | --- |

| Overall coordination | Directs strategy and makes key decisions. |

| Communication | Drafts messages, prepares clarifications and manages spokespeople. |

| Verification | Cross-checks information, analyses the falsehood and assesses its actual reach. |

| Legal | Assesses legal implications and institutional risks. |

| Technology/data | Monitors networks, analyses metrics and detects trends. |

| Relevant services | Provide technical information and operational support. |

| Citizen services | Answer queries and channel citizens’ concerns. |



| Hour | Stage | Tasks displayed |

| ---: | --- | --- |

| 0 | Detect viral content | Alert triggered by an unusual increase in mentions; identify the falsehood and make an initial assessment; activate the crisis room immediately. |

| 1 | Internal verification | Coordination meeting; gather data and background; confirm facts within the institution. |

| 2 | Consult official sources | Contact relevant departments and bodies; request official information and documents; cross-check data. |

| 3 | Response decision | Assess response options; approve the key message and spokespeople; communication and channel plan. |

| 4 | Publish clarification | Disseminate the official message; publish through institutional channels; coordinate with media and allies. |

| 5 | Monitor impact | Real-time monitoring; analysis of metrics and sentiment; detect new related falsehoods. |

| 6 | Report and learning | Evaluate the response and results; identify successes and areas for improvement; update protocols and prevention plans. |



**Critical decisions:** confirm whether the falsehood is false, misleading or out of context; determine actual impact and potential reach; decide the best time to respond; define a clear, truthful and empathetic key message; choose spokespeople and communication channels; maintain constant internal coordination.



**Mistakes to avoid:** responding without checking information; minimising or ignoring the falsehood; contradictions between departments or spokespeople; technical or defensive language; attacking or disparaging citizens; publishing late or through unsuitable channels; failing to document decisions and lessons.



The central screens show “Network monitoring: 32.7K total mentions”, sources X (Twitter), Facebook, news and others; “Viral falsehood detected — high potential reach — verifying information”; impact and development lines labelled mentions/potential reach; and sentiment values of 62% negative and 24% neutral. A positive category is visible, but its value is obscured. Small source-count labels and graph ticks are not reliably legible. Key messages: transparency, verified information, commitment to truth, service to citizens. The room sign says “Institutional coordination”; the flipchart lists truthfulness, speed, responsibility, empathy and public service.



Footer: “Truthfulness · speed · coordination · empathy · public service. Informing people well, on time and responsibly strengthens public trust.”



### A visual kit for a citizens’ assembly, as an example



*Prompt, translated from Spanish:*



> Create a visual kit for a citizens’ assembly on “How should our council use artificial intelligence?” The image should look like a deliberative working table viewed from above, with carefully organised printed materials. Include: 1. A card with debate rules: listen; argue respectfully; ask for evidence; distinguish facts from opinions. 2. Three scenario cards: AI for citizen services; AI for internal management; AI for security and emergencies. 3. A risks card: privacy; bias; errors; digital exclusion. 4. An opportunities card: speed; accessibility; time savings; better services. 5. A final sheet: “Questions for deciding”. Style: realistic, warm, institutional and participatory, like materials prepared by a council for a citizens’ event. Brief, legible Spanish text. Format 16:9.



[![Generated citizens' assembly materials on council use of AI, arranged on a table.](https://elcontemplador.github.io/estrategia-english/assets/images/3a2bccb3-442a-4f6f-b54f-f2d7842b3e72_1672x941.png)](https://substackcdn.com/image/fetch/$s_!wXID!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3a2bccb3-442a-4f6f-b54f-f2d7842b3e72_1672x941.png)



*English transcription of the generated kit: “Citizens’ assembly — How should our council use artificial intelligence? We participate today; we decide better.”*



**1. Debate rules:** listen to every opinion without interrupting; argue respectfully, without personal attacks; ask for data and examples before expressing an opinion; distinguish facts from opinions, separating what we know from what we think. “Every voice counts.”



**2. Scenario 1 — AI for citizen services:** chatbots and assistants 24/7; clear, personalised information; faster replies; accessible channels for everyone. “Which uses would we prioritise? What precautions should we take?”



**3. Scenario 2 — AI for internal management:** automate repetitive tasks; analyse data to make better decisions; detect incidents and optimise resources; support municipal staff. “Which processes would we improve? Where should we set limits?”



**4. Scenario 3 — AI for security and emergencies:** early risk detection; more effective coordination; real-time information; prediction and prevention. “Which benefits and precautions matter most?”



**5. Risks:** privacy — protect personal data and ensure responsible use; bias — avoid discriminatory or unfair algorithmic decisions; errors — AI can fail, so we need human oversight; digital exclusion — nobody should be left behind through lack of access or skills. “Fewer risks, more trust.”



**6. Opportunities:** speed — faster replies and procedures; accessibility — services that are easier for everyone to use; time savings — less waiting, more time for what matters; better services — data-based decisions to improve life in the city. “Use technology for a fairer and more efficient city.”



**7. Questions for deciding:** What problems in our municipality could AI help solve? Which AI uses do we consider priorities, and why? Which risks concern us most, and how can we mitigate them? What safeguards and limits must the council observe? What do we need to ensure nobody is left behind?



Additional cards say: “Your participation improves our municipality. Thank you.” “Your opinion builds our city.”



### A stakeholder map, in this case for a low-emission zone



*Prompt, translated from Spanish:*



> Act as a political communication strategist and stakeholder-map designer. Create a square infographic entitled “Stakeholder map for a new low-emission zone”. Visually represent five groups: 1. Affected residents. 2. Businesses and hospitality. 3. Environmental associations. 4. Regular motorists. 5. Municipal government and opposition. For each group include: likely attitude — support, uncertainty, opposition or negotiation; main concern; recommended message; most useful communication channel. Organise the composition as a map of connected nodes around the central public measure, “Low-emission zone”. Style: institutional, analytical and modern; no caricatures; no partisan propaganda; clear icons; perfectly legible Spanish text; a sober palette, like a strategic report. Format 1:1.



[![Generated stakeholder map around a proposed low-emission zone.](https://elcontemplador.github.io/estrategia-english/assets/images/5f2ce96b-39b4-407a-85f5-bf344098dad5_1254x1254.png)](https://substackcdn.com/image/fetch/$s_!hg9X!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F5f2ce96b-39b4-407a-85f5-bf344098dad5_1254x1254.png)



*English transcription of the generated stakeholder map: “Understand each stakeholder to design effective and fair implementation”. Central objective: improve air quality, health and quality of life in the city.*



| Group | Likely attitude | Main concern | Recommended message | Most useful channel |

| --- | --- | --- | --- | --- |

| Affected residents | Uncertainty | Mobility restrictions and possible financial costs; unequal effects across neighbourhoods. | Less pollution, better health and quality public spaces for everyone. | Residents’ meetings, neighbourhood associations, local social media and the council website. |

| Businesses and hospitality | Negotiation | Possible loss of customers and sales; access difficulties and loading logistics. | A more attractive environment, customers who walk and stay, support and guidance. | Business associations, sector meetings, email and WhatsApp Business, signs in shopping areas. |

| Environmental associations | Support | The measure may be insufficient or have too many exceptions. | Health and climate first; clear ambition, a realistic timetable and transparent evaluation. | Technical working groups, public reports, press conferences and social media. |

| Regular motorists | Opposition | Access restrictions, fines and lack of affordable alternatives. | Real alternatives to cars, a gradual timetable and help to adapt. | Local media, radio, SMS/WhatsApp, urban signs and the council website. |

| Municipal government and opposition | Negotiation | Political cost, social consensus and the measure’s perceived effectiveness. | A necessary, fair, evidence-based policy; dialogue and continuous evaluation. | Council plenaries, committees, appearances, technical reports and institutional communication. |



**Principles across all groups:** active listening and real participation from the outset; clear data and visible benefits for the city; support measures and a just transition to help people adapt; transparency, evaluation and continuous adjustments; simple, consistent and relatable messages.



### A rapid briefing format: a decision in five minutes, for advising politicians or public managers



*Prompt, translated from Spanish:*



> Act as an adviser in a public-sector cabinet and a strategic information designer. Create a portrait A4 visual briefing entitled “Decision in five minutes: introducing generative AI into municipal citizen services”. The piece should be designed for a local government meeting. Include these clearly differentiated blocks: 1. Public problem. 2. Objective of the measure. 3. Three decision options: limited pilot; gradual deployment; broad deployment. 4. Expected benefits. 5. Administrative risks. 6. Ethical and legal risks. 7. Minimum conditions for approving the measure. 8. Final recommendation. Very important: do not invent figures. If information is missing, use placeholders such as “[data pending]”. Text must be brief, perfectly legible and in Spanish. The design should resemble an executive document from a European public-sector consultancy: sober, clear and elegant, with discreet icons, a strong visual hierarchy and plenty of white space. Portrait A4 format.



[![Generated executive briefing on introducing generative AI into municipal citizen services.](https://elcontemplador.github.io/estrategia-english/assets/images/94a94ea7-0ab3-47f7-91ac-2b8f514bd129_1024x1536.png)](https://substackcdn.com/image/fetch/$s_!MHyg!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F94a94ea7-0ab3-47f7-91ac-2b8f514bd129_1024x1536.png)



*English transcription of the generated briefing: “Decision in five minutes: introducing generative AI into municipal citizen services — Briefing for a local government meeting”. Date: [data pending].*



**1. Public problem:** high demand across multiple channels — telephone, website, email, in person and social media; slow and uneven responses depending on channel or time; a heavy administrative burden on municipal staff; information scattered across websites and complex procedures. Impact: lower citizen satisfaction and reduced institutional efficiency.



**2. Objective:** provide faster, clearer and more consistent answers 24/7; increase service capacity without expanding the organisational structure; free staff time for higher-value tasks; improve citizens’ experience and trust in the council. Approach: generative AI supports municipal staff rather than replacing them.



**3. Decision options:**



| Option | Scope and requirements | Assessment displayed |

| --- | --- | --- |

| A. Limited pilot | Scope: [specific area or service], e.g. general information and frequent procedures. Channels: web and chat. Duration: [X months]. Evaluation: key indicators and learning. | Ideal for learning, validation and adjusting risks on a small scale. |

| B. Gradual deployment | Expansion by priority areas and use cases; integration with systems and knowledge bases; staff training and support; continuous evaluation and adjustments. | Balances impact, learning and change management. |

| C. Broad deployment | Coverage across channels and departments; AI integrated into key citizen-service processes; advanced personalisation and automation; established data and model governance. | Maximum impact and efficiency; requires organisational and technological maturity. |



**4. Expected benefits:** availability 24/7 and faster responses; greater service capacity without increasing structural costs; more consistent answers based on official information; time freed for staff to handle complex tasks and provide support; improved citizen experience and satisfaction.



**5. Administrative risks:** technological and supplier dependence; complex integration with existing systems; information quality and updating; change management and staff adoption; medium- and long-term financial sustainability [data pending].



**6. Ethical and legal risks:** personal-data protection and privacy; algorithmic bias and indirect discrimination; lack of transparency and explainability; compliance with the GDPR and EU AI rules [data pending].



**7. Minimum approval conditions:** data protection impact assessment; AI governance framework and a designated person responsible; official, traceable knowledge base; human oversight and escalation protocol; transparency to citizens about AI use; staff training and support plan; continuous evaluation and clear indicators.



**8. Final recommendation:** begin with option A, a limited pilot in a priority service, with rigorous evaluation and minimum conditions from the outset. If results are positive, move towards option B, gradual deployment, preparing capabilities, integration and governance to scale safely, ethically and sustainably.



**Proposed next step:** approve the pilot design and governance framework; assign responsibilities and a timetable [proposed date pending]. Note in the image: “This information is indicative. Detailed analysis and specific data are required for a final decision.”



### Analysis of hairstyles or recommended clothing colours



One of the more “curious” uses, and one enjoying considerable success online, applies these models and their multimodal capabilities to questions of style and personal appearance. Here are a couple of prompts and examples — this time [we have used as our model the AI “candidate” we created for the European elections back in April 2024, in issue 29 of estrategIA](https://open.substack.com/pub/estrategiabyaleph/p/estrategia-29-inteligencia-artificial?utm_campaign=post-expanded-share&utm_medium=web). I am not a specialist in this field and do not have sufficient expertise to judge the results in depth. They will probably depend heavily on the reference photograph itself and how it was taken. But this seems an interesting advance on the simpler hairstyle trial we presented in the December issue.



*Prompt, translated from Spanish:*



> Create a hairstyle-analysis graphic using this portrait. Show side-by-side hairstyle comparisons to highlight which ones best suit the person portrayed.

>

> Prioritise the visual element, using only short labels and no paragraphs.



[![Generated hairstyle comparison for the fictional AI candidate, with recommendations transcribed below.](https://elcontemplador.github.io/estrategia-english/assets/images/2514c74f-7f4b-4a10-8597-dddc4a4b0933_1402x1122.png)](https://substackcdn.com/image/fetch/$s_!gg-M!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2514c74f-7f4b-4a10-8597-dddc4a4b0933_1402x1122.png)



*English transcription of the model’s hairstyle analysis, using the fictional candidate: “Find your best version”.*



**Starting point:** heart-shaped face; straight or wavy hair; medium-to-high density. **Most flattering:** long layers, textured lob, feathered hair. **May also work:** classic bob, medium-length layers, centre parting. **Less recommended:** straight cut without layers.



**Flattering features:** layers that add movement; subtle highlights for brightness; volume through the mid-lengths and ends; a slightly side-set parting. **Details that make a difference:** soft highlights, face-framing layers, side parting, natural waves and volume at the roots.



*Second prompt, translated from Spanish:*



> Using this portrait, create a personal colour analysis centred on diagrams. Show which clothing colours best suit the person through visual comparison. Keep text to a minimum and avoid paragraphs.



[![Generated personal colour analysis for the fictional candidate, with palette and combinations described below.](https://elcontemplador.github.io/estrategia-english/assets/images/a8475cc6-e245-48d0-9ed9-ed8ca16833c1_1024x1536.png)](https://substackcdn.com/image/fetch/$s_!dCFY!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa8475cc6-e245-48d0-9ed9-ed8ca16833c1_1024x1536.png)



*English transcription and description of the generated personal colour analysis. These are the image’s recommendations, which the author explicitly says he cannot assess in depth.*



“Your palette”: undertone scale cool–neutral–warm, with the marker towards warm; depth light–medium–deep, with the marker towards deep; intensity soft–medium–high, with the marker around medium. Overall temperature: warm-neutral. A high-contrast black-and-white circle is selected among four contrast samples.



**Flattering colours:** portraits in dark teal, olive, rust, cream and navy. The fifteen swatches include cream, beige, tan, ochre-brown, dark brown, olive tones, green, teal, navy, plum and warm reds/oranges. **Less flattering colours:** portraits in pale pink, lilac, pale blue and grey; swatches in pale pink, lilac, blue-grey, pale blue, grey, muted blues, charcoal and black.



**Ideal neutrals:** cream, beige, taupe, brown and dark blue-grey. **Metallics:** gold and copper. **Recommended patterns:** leopard, paisley and houndstooth. **Combinations that work:** dark green/beige/ochre; navy/olive/cream; rust/cream/brown; teal/brown/ochre. **Avoid very cool or neon outfits:** the crossed-out combinations are neon pink/blue/yellow-green; pale aqua/lilac/pink; and pale grey/slate blue/navy.



As these examples show, ChatGPT Images 2.0 represents a significant leap in AI’s visual generation and editing capabilities; the leap is functional, not aesthetic. It still has limitations — for example, it still fails to depict certain Salamanca monuments accurately or place them correctly on a map — but it opens up a very promising field of experimentation for public, political and institutional communication. The underlying consequence is clear: the ability to produce high-quality visual communication ceases to be a bottleneck for small institutions, which may help balance communication against major political actors. The decisive question is no longer simply what the model can generate, but what responsible, clear and strategic uses we can design with it.



[Fernando Nieto Lobato](https://www.linkedin.com/in/fernandonietolobato/)



*Director of Digital Innovation at Institución Educativa ALEPH and editor of the estrategIA newsletter*