Prompting in Studio
How Studio handles your prompt, how to pick a model for the job from its stated strengths, and how to test cheaply before you spend on an expensive generation.
For the complete documentation index, see llms.txt. This page is also available as Markdown.
#What Studio does with your prompt
- Studio sends your image prompt to the model exactly as you typed it. It does not add a style suffix or strip text instructions.
- The model you pick is the model that runs. If it fails, nothing else is tried in its place and the charge is refunded.
- Expand prompt opens a larger editor for long prompts. Clear prompt empties the field.
- The Lock-in agent is different: it can ask you questions and write prompts for you. See The Lock-in agent.
#Pick the model for the job
Each model's page states what it is built for. These strengths are the ones the model list describes:
| You want | Models described for it |
|---|---|
| Text inside the image, posters, typography | GPT Image 2, Ideogram V4, Ideogram 3, Qwen-Image 3.0 |
| Logos and vector-style graphics | Recraft V4 |
| Photorealism | FLUX 2 Pro, GPT Image 2, FLUX 2 Max (maximum fidelity) |
| Premium detail and realism | Nano Banana Pro |
| Editorial looks | Seedream 5 Pro |
| Low-price drafts for iteration | Z-Image Turbo (6 to 11 credits, by image shape), FLUX Schnell (7), Recraft V4.1 Flash (13) |
#Test cheap, then spend
Step 1: Draft on a low-price model
Run your prompt on Z-Image Turbo or FLUX Schnell. A draft costs 6 to 11 credits and shows whether the composition works. Models that do not take a reference image cannot test a reference, so draft with a model that does if you plan to use one.
Step 2: Fix the prompt, not the model
Change one thing at a time: subject, setting, lighting, style. Compare the cards side by side.
Step 3: Run the final on the model you want
Switch to the premium model once the prompt reads well. Use Count to get up to 4 options in one run.
#Prompt writing that helps with any model
- Name the subject first, then where it is, then the light and the camera or style.
- Say what you want, not only what you do not want.
- Put exact text you need in the image in quotes, and use a model described for text rendering.
- Use a reference image when the look is easier to show than to describe. It is one image per run.
#Video prompts
Describe the scene and what moves. Length is a setting, not a prompt instruction, and it drives the price. Test at a short length, then extend. With a start frame, the video begins from your image. See Create a video.
#Acceptable use
The Terms prohibit deepfakes, non-consensual intimate imagery, minors in harmful contexts, impersonation of real individuals and misleading political content. UOVA reserves the right to monitor prompts and outputs to prevent misuse. See the Terms of service.
#Common questions
The Privacy Policy states that UOVA does not use your prompts or generated content to train its own AI models, and configures its AI providers not to train on your inputs where such controls exist. Third-party providers may retain inputs temporarily for abuse monitoring.
The Terms grant a license to use AI-generated content created through your account for personal and commercial purposes, including websites, marketing materials and social media. They make no representation about copyright ownership of AI content, and outputs are not guaranteed to be unique. See Data, security and IP.