Prompt extractor · Flux and Stable Diffusion
Image to prompt for Flux and Stable Diffusion
Last updated
First check whether the PNG still has its metadata. Stable Diffusion and ComfyUI often save the exact prompt inside the file. If it is empty, extract a prompt from the image itself: write it as a plain caption for Flux, or as weighted tags with a separate negative prompt for SD 1.5 and SDXL.
Want to skip ahead? The free prompt extractor reads any JPEG, PNG, or WebP and gives you a text prompt and a JSON prompt. No account needed.
Can you read the exact prompt from the file?
Often, yes, and it is worth the 20 seconds. An untouched PNG from Automatic1111 or Forge carries the positive prompt, negative prompt, seed, sampler, and steps. A ComfyUI PNG carries the full workflow. Open it in PNG Info, drop it on the ComfyUI canvas, or use a reader. Our comparison of MaPic, SD Prompt Reader, and ExifTool covers the free options.
If the file came from Reddit, Discord, or a JPEG export, the text chunk is almost always gone. Move on to extraction.
What does an extracted Stable Diffusion prompt look like?
We ran the photo above through the Stable Diffusion prompt generator on 2026-10-07, set to SDXL with a photoreal checkpoint in mind. It uses the same vision step as the prompt extractor and then formats for SD. Unedited output:
Positive prompt
(three hands clinking coffee cups:1.3), (top down view:1.2), holding cups, latte art, iced coffee, dark liquid, coffee shop table, wooden surface, dark ceramic saucers, silver spoons, bokeh background, soft natural lighting, (photorealistic:1.2), macro photography, close up shot, neutral tones, warm atmosphere, high contrast, indoor setting, blurred background, realistic textures, detailed foam, sharp focus on cups, composition from above, hands reaching inwards
Negative prompt
distorted fingers, extra hands, fused fingers, morphing, deformed hands, missing fingers, bad anatomy, blur, low quality, watermarks, signatures, text, jpeg artifacts, overexposed, washed out colors, messy, unnatural lighting, warped cups
The negative prompt leans hard on hands, which makes sense for a picture that is mostly fingers and cups. Suggested settings came back as 1216 by 832, DPM++ 2M Karras, 30 (25 to 40) steps, CFG 6 (5 to 7).
And the same image as a Flux prompt?
Flux reads a description the way a person would. Weights in brackets do little, and there is no negative field in most Flux setups. The same run produced this caption, which is what we would paste into Flux:
Flux caption
A high-angle close-up shot shows three hands clinking coffee cups together in a cafe. Two of the cups contain lattes with intricate heart-shaped foam art, while the third cup holds iced black coffee. The hands are positioned over a wooden table with dark ceramic saucers visible below.
If the result looks too clean, add the camera details back in plain words: "shot from above on a 35mm lens, shallow depth of field, soft window light." Flux tends to follow those literally.
Which visual details matter most?
- Subject and action. Who or what, doing what. Put it first.
- Camera angle and lens. "Top down" versus "eye level" changes everything else in the frame.
- Light. Direction and quality: soft window light, hard noon sun, neon rim light.
- Palette and medium. Photo, oil paint, 3D render, plus two or three colors.
- What to avoid. For SD only, put this in the negative prompt instead of the positive one.
Questions people ask
Do Flux and Stable Diffusion need different prompts?
Yes. Flux understands full sentences well and has no real use for a negative prompt, so a plain caption works best. Stable Diffusion 1.5 and SDXL respond well to comma-separated tags, weights like (subject:1.2), and a separate negative prompt.
How do I read the prompt from a Stable Diffusion PNG?
Automatic1111 and Forge save the prompt and settings in a PNG text chunk called parameters. Open the PNG Info tab in the web UI and drop the file in. ComfyUI saves the whole workflow, so dragging the PNG onto the ComfyUI canvas loads the graph with the prompt inside. Metadata readers like SD Prompt Reader do the same outside the app.
Why is the metadata empty?
The image was probably saved as JPEG, edited, or passed through a site that strips metadata. Discord, Reddit, Instagram, and most image hosts do this. Once the text chunk is gone, you have to extract a new prompt from the pixels.
Which settings should I start with?
For SDXL, stay near one megapixel (for example 1216 by 832 for a 3:2 image), around 30 steps, and CFG around 6 with DPM++ 2M Karras. For Flux dev, use a guidance of about 3.5 and skip the negative prompt.