Create your first image in 30 seconds
- Select Image Generation on the home page.
- Choose a model. The panel automatically shows only its supported sizes, ratios, formats, and generation controls.
- Enter a prompt. For an image-edit model, upload clear references and wait for every upload to finish.
- When available, choose prompt optimization, a random seed, or smart prompt expansion.
- Review the live credit estimate and select the Generate button on the right.
- In My Images, preview, download, share, reuse settings, or open the asset manager.
For a first attempt, keep the default model and size. Describe the subject, setting, and style clearly before adding smaller details.
A practical prompt formula
Use: subject + action/state + setting + composition + lighting + color + style + essential details.
Example: A young woman in a white trench coat standing outside a convenience store on a rainy night, holding a transparent umbrella, medium shot, subject on the right, blue and violet neon reflected on wet pavement, cinematic lighting, shallow depth of field, crisp detail.
Useful habits:
- Put the most important subject first and replace vague words such as “beautiful” with visible details.
- Specify framing: close-up, medium shot, full body, overhead, low angle, symmetrical, or rule of thirds.
- Specify lighting: soft window light, backlight, neon, golden hour, or high contrast.
- For text in an image, provide the exact short copy, placement, color, and desired type style.
- Change only one or two variables per retry so you can tell what improved the result.
Text generation and reference editing
Use a text-to-image model when creating from scratch. Use a model marked Edit when identity, product appearance, or the main composition should come from a reference image. For edits, state what must remain, what must change, and the target look.
Example: Preserve the face and hairstyle from the reference. Change the outfit to a black evening gown and the background to a gold theater lobby. Warm overhead light, full-body composition, realistic cinematic finish.
Reference limits vary by edit model: Qwen Image Edit Spicy accepts 1 image, Wan 2.6 Edit up to 4, Wan 2.7 Edit up to 9, and Seedream v5.0 Pro Edit up to 10. Each image can be up to 10 MB. Follow the current limit shown in the uploader.
Size, ratio, and advanced controls
- Size and ratio: Seedream and Wan 2.6 expose pixel dimensions. Wan 2.7 combines 1K/2K/4K tiers with aspect ratios. Options vary by model.
- Output format: supported models let you choose JPEG or PNG.
- Generation mode: Optimize Prompt enriches the request; Do Not Optimize follows an already detailed prompt more literally.
- Random seed: Spicy image models accept a non-negative integer. Reusing a seed with similar settings can help reproduce a direction; 0 means random.
- Smart prompt expansion: Z-Image Spicy can automatically enrich prompt details.
After a model change, controls that do not apply disappear and require no setting.
Image model guide
| Model | Best for | Tip |
|---|---|---|
| Seedream v5.0 Pro | Detailed characters, posters, scenes, and complex directions | A strong general starting point; describe composition and light |
| Seedream v5.0 Pro Edit | Reference-based transformations and identity/product consistency | Explicitly separate what to preserve from what to change |
| Wan 2.7 | General creation at 1K or 2K | Useful for exploring formats and creative directions |
| Wan 2.7 Pro | Higher-resolution output up to 4K | Use after the composition is settled |
| Wan 2.7 Edit | Reference-based image changes | Start with a clear primary reference and avoid conflicting edits |
| Wan 2.6 Text/Edit | Multi-format creation with exact pixel sizes | Choose it when a specific output dimension matters |
| Z-Image Spicy | Multi-ratio text-to-image creation | Supports a random seed and optional smart prompt expansion |
| Qwen Image Edit Spicy | Single-reference image editing | Upload one primary reference and separate preserved and changed details |
Available models, defaults, and prices may be updated by the platform. Always use the options currently shown in the creation panel.
Troubleshooting
- Character identity changes: reuse one clear reference and keep age, hair, outfit, and style descriptions consistent.
- Crowded or misplaced subjects: reduce the number of subjects and state their positions explicitly.
- Unexpected output: shorten the prompt to its key constraints, establish composition first, then add material, lighting, and mood.
- Generate is unavailable: check required uploads, the 10 MB image limit, model/size compatibility, and available credits.
