This checklist covers preparing an Image to Image task in the image generator. For the full generation workflow, use the getting-started guide.
Check the reference file before uploading
Sign in, select Image to Image, then choose a model available for that mode. Z-Image Turbo itself is currently a text-to-image option, so reference editing requires another available model and uses credits. Use a reference you have permission to upload and edit, including permission to use a person's photo where applicable. Do not assume a picture is available for reuse just because it appears online.
- Use a genuine JPEG, PNG or WebP image. Changing a filename extension does not convert a file; the upload service also checks the image content.
- The current generator shows a 5 MB per-image limit. Follow the limit shown in the upload control and reduce the file size if needed.
- Check the selected files and the number allowed by the current control. Flux 2 Klein 9B currently allows up to 4 references; other currently offered image-to-image models allow up to 9. Switching models can lower that limit, so remove extra files before proceeding. A file appearing as a preview does not by itself mean its upload completed.
The file picker may show other image formats. That does not mean every format is accepted by the upload service or every model. If a file is rejected, export a supported image from your image editor rather than renaming it.
Wait for the upload to finish
Wait until the reference finishes uploading before generating. If an upload fails, remove that failed item and add the corrected file again. Review all remaining references so an older or unintended image is not used.
Selecting references and submitting a generation are separate steps. Do not repeatedly press Generate to fix an upload error. Check the prompt, model and displayed credit cost after the references are ready. See the free and credit guide for the current per-task rules.
Keep the prompt within the displayed limit
The generator checks a 2,000-character prompt limit using its on-page counter. Describe what to retain from the reference and what to change. For example, you can describe a new background or lighting while identifying the subject you want to keep.
That is a way to structure instructions, not a promise of exact identity, layout or text preservation. Review the model, aspect ratio and cost before choosing Generate.
Three ways to describe “keep” and “change”
Start with one clear reference that shows the detail you need. The following are suggested editing instructions, not tested results. Models can interpret the same instruction differently, and none of these examples guarantees an exact face, room layout, product shape or label.
Portrait: keep the pose, change the background
Use a portrait you are allowed to edit, with the face and shoulders clearly visible. For a direction similar to courtyard portrait, try:
Retain the person's pose, hairstyle and light-blue shirt from the reference. Replace the background with a simple green courtyard in soft morning light. Frame from the chest up, with the entire head visible and natural skin texture.
If the background changes too little, name the replacement clearly instead of adding several unrelated styles. If the face or clothing changes unexpectedly, simplify the request to one background change and compare those details again. Do not treat an output as verified identity preservation.
Interior: keep the layout, change one material
Use a room photo that clearly shows the window, shelves and furniture. For the sunlit reading room direction, try:
Keep the window position, camera viewpoint, armchair and table placement from the reference. Change only the bookshelf finish to light oak. Keep the walls cream-colored and the vertical lines straight; do not add furniture.
If the room is rearranged, remove other requested changes and repeat the specific positions to retain. Compare the window and furniture locations before trying a second change, such as warmer afternoon light. This is a visual edit, not a dimensionally reliable renovation plan.
Still life: keep the bottle, change the setting
Use a clear photo of a bottle you have permission to use, with the cap and base fully visible. For the amber fragrance direction, try:
Retain the bottle silhouette, amber glass color and matte cap from the reference. Place the bottle on a cream stone pedestal against a plain beige background, with soft diffused side light. Keep the whole bottle visible; do not add another bottle or decorative props.
If the bottle shape changes, simplify the edit to the background only. Check the outline, cap, proportions and any label separately. Do not use generated packaging or lettering as an accurate product record without checking it against the original.
The links above load the Inspiration text prompt only, in the current language; they do not attach a reference, switch to image-to-image, select a model or generate. After opening one, choose Image to Image, choose a model, upload your own permitted reference and replace or adapt the text for the specific edit. Review the ratio and credit cost, then explicitly choose Generate. A restored login draft may take priority over a linked prompt, so check the text too.
What the sign-in draft can recover
When the supported sign-in flow interrupts your input, a temporary draft can retain your prompt, mode, model and aspect ratio. It is intended for returning in the same tab to the same generator entry and language within 15 minutes.
The draft is not a saved project. It does not preserve reference files, generated images or task history. Add your references again after signing in, then inspect the restored options and explicitly submit when ready. Restoring input does not automatically generate an image.
If the draft is missing, check whether you changed entry page or language, waited too long, edited the input already, or used a browser that blocks temporary storage. A model that is no longer selectable cannot be restored. Keep your own copy of important prompt text; recovery is not guaranteed across tabs, browsers or devices.
Ready to generate, or already submitted?
If you have not submitted, return to the generator and check the upload, prompt and cost once more. If a task was already accepted, check your image tasks before creating another one.
For pending tasks, missing results and download problems, follow the task and download troubleshooting checklist. A new generation is a separate task and may use credits.