AI Image Generation Workflow with AgentEarth
Use natural language to make an AI agent generate posters, illustrations and concept art, compare the same prompt across multiple text-to-image models, and refine the winner in place.

Tool names in this guide are for reference only and may have changed or been renamed. Please use the latest tools in the tool marketplace when implementing.
Key takeaways
- Generate posters, illustrations and concepts from one prompt, comparing models
- Call every text-to-image model from one platform and key
- Hand the output straight into editing or background removal
Tools used
Why getting one good image is such a hassle
The hard part of AI image generation usually isn't writing the prompt — it's that the tools are scattered. Models differ a lot in style: one is photorealistic and detailed, another leans illustrative, another is fast but rougher. To compare them you'd sign up for each service separately, learn separate parameters, pay separately — and only after all that discover one doesn't fit, so you start over. Then you want to tweak one small detail and that's yet another tool. The whole thing is dragged down by switching costs, and the actual creative part ends up being the fastest step.
An AI agent connected to image tools folds this into one entry point: you describe what you want in natural language, and the agent picks the model, generates, compares, and even refines — one instruction at a time.
The workflow in four steps
- Describe the subject, style, size and aspect ratio in natural language.
- Generate a first draft with a text-to-image tool.
- Compare the same prompt across models and pick the best fit.
- Refine the chosen image locally, and output the final image plus a reusable prompt.
How to do each step
Step 1: Describe the image clearly
Tell the agent in plain language what you want: the subject, the style (photoreal / illustration / flat), the size and aspect ratio, the mood. The more specific the description, the closer the first draft.
use ae to generate a 16:9 poster of a mountain lake at sunrise, flat illustration style, calm mood
Step 2: Generate a first draft
The agent calls a text-to-image tool to produce a draft. For high-quality general output use ae_openai_gpt_image_2_5; for fast drafts use ae_nano_banana_2.
use ae to create a first draft with ae_nano_banana_2
Step 3: Compare across models and pick
Run the same prompt through several models and compare style and quality before choosing. Beyond the two above, you can add an alternate-style model like ae_wan_27_image_pro_text_to_image. This is the point of one platform with many models: no separate sign-ups, same sentence, direct comparison.
| Goal | Models to try (examples) | Good for |
|---|---|---|
| High-quality general | ae_openai_gpt_image_2_5 |
Quality-sensitive, detailed work |
| Fast draft | ae_nano_banana_2 |
Rapid iteration, generating several options |
| Alternate style | ae_wan_27_image_pro_text_to_image |
A different stylistic take for contrast |
Step 4: Refine and keep the prompt
Once you pick a version, use an image-editing tool like ae_nano_banana_2_edit to adjust a region instead of regenerating the whole thing. Finally, have the agent output the final image together with the prompt you used, so you can reuse it next time.
Single model vs one platform with many models:
| One model only | Many models on one platform |
|---|---|
| Style limited to one model | Covers photoreal to illustration |
| Switch platforms if it doesn't fit | Swap models with the same sentence |
| Find another tool to edit afterwards | Refine in place right after generating |
How to get started
- Sign up — new users get 1,000 free credits.
- Connect in-app from an agent like WorkBuddy or Trae by searching for "AgentEarth", or paste a one-line MCP config into Claude Code, Cursor or VSCode.
- Describe the image in natural language and let the agent pick, generate, compare and refine.
New to MCP? Start with What is MCP?; for connection details see the access guide.
FAQ
Can I control size and aspect ratio? Yes. State the ratio (e.g. 16:9) and intended use in your description, and the agent passes those constraints to the text-to-image tool.
Which model is best for photoreal vs illustration? There's no single answer — which is why running the same prompt across several models is recommended. Start with ae_openai_gpt_image_2_5 for general output and ae_nano_banana_2 for fast iteration, then add an alternate-style model for contrast.
Can the generated image be edited? Yes. After choosing a final image you can make local tweaks with an editing tool — see the consistent image editing guide.
