Turn your image into a Stable Diffusion prompt
Upload a picture and get clear, literal prompts for Stable Diffusion, SDXL and Flux, plus a title and keywords. Free, no sign-up needed.
Drop your image here or click to upload
PNG, JPG, GIF or WebP · up to 50 MB each
Prompts for Stable Diffusion, SDXL and Flux
Stable Diffusion models follow detailed, descriptive prompts well. ph-imetix writes prompts you can paste into AUTOMATIC1111, ComfyUI, Forge, Fooocus, InvokeAI or any online Stable Diffusion or Flux generator. The Direct tone is pre-selected because concise, literal wording gives SD models the clearest instructions.
Turning the prompt into a great image
- Paste the generated prompt into the positive prompt box.
- Add a negative prompt (see the starters below).
- Pick a checkpoint that suits the style: a photoreal model for photos, an anime model for illustrations.
- Set the resolution your model was trained for. SDXL works best around 1024 px, SD 1.5 around 512–768 px.
- Generate a batch, choose the best, then refine with img2img or inpainting.
Negative prompt starters
| Image type | Negative prompt to start with |
|---|---|
| Photos & portraits | blurry, lowres, deformed hands, extra fingers, bad anatomy, watermark, text |
| Illustrations | photo, realistic, blurry, jpeg artifacts, watermark, signature |
| Products | clutter, dust, scratches, distorted logo, text, watermark, low quality |
| Landscapes | people, buildings, text, oversaturated, blurry, frame |
Settings that matter
| Setting | What it does | Typical starting point |
|---|---|---|
| CFG scale | How strictly the image follows the prompt | 5–7 for most models |
| Steps | How many refinement passes | 20–30 |
| Sampler | The method used to create the image | Your interface default is fine to start |
| Seed | Repeat or vary a result | Fix it while you edit the prompt |
Exact best values depend on the model; check the model card of your checkpoint.
Writing for Flux
- Keep the prompt in natural sentences. Flux understands long descriptions well.
- Be precise about positions and counts (“two cups on the left”).
- Put any text you want in the image in quotation marks.
- Describe what you want instead of relying on a negative prompt.
See how Flux differs from SD and Midjourney.
Copying a style you like
Upload a reference image, generate 5 variants, then take the style words that repeat across them, such as the medium, lighting and palette. Those words are the “style signature” you can reuse with any subject, and they combine well with a style LoRA if you use one.
Using prompts with img2img and ControlNet
- img2img: load your reference, paste the generated prompt, and use a medium denoising strength to restyle it.
- ControlNet: keep the pose or edges of your image while the prompt sets the style and details.
- Inpainting: use part of the prompt to regenerate only one area.
Related guides
Midjourney vs Stable Diffusion vs Flux vs ChatGPT: How to Write Prompts for Each — how each generator reads prompts, and how to adapt one prompt for all four.
10 Tips to Write Better AI Image Prompts (With Examples) — before-and-after examples for every major generator.
AI Product Photography Prompts: Turn One Photo into Many Scenes — prompt formulas for product scenes, lifestyle shots and ad backgrounds.
Frequently asked questions
Does it write negative prompts?
It writes the positive prompt. Use the negative prompt starters on this page as a base.
Do the prompts work with Flux?
Yes. Flux responds well to the descriptive, sentence-style prompts ph-imetix produces.
Should I use SD 1.5 or SDXL?
SDXL generally gives better detail at 1024 px; SD 1.5 has a huge library of community models. The prompts work with both.
Can I use it for img2img?
Yes. Use the prompt together with your image in img2img or ControlNet to guide the result.
Which tone is best for Stable Diffusion?
Direct is pre-selected for literal, concise prompts. Try Vivid for artistic checkpoints.
Can I get the prompt as keywords instead of sentences?
Use the keyword list (5–50) as tags, or choose the Simplified tone for shorter prompts.