Remove Stable Diffusion Metadata from Images
Stable Diffusion interfaces like Automatic1111, ComfyUI, and Forge embed detailed generation parameters into every image. This includes your full prompt, negative prompt, seed, CFG scale, steps, sampler, and model hash. StripShot removes all of it.
What Stable Diffusion embeds in your images
Stable Diffusion itself is an open-source model. The metadata comes from the interfaces used to run it. Each interface stores generation data differently, but most use PNG tEXt chunks and EXIF fields. Here is what each interface embeds.
PNG tEXt chunks (Automatic1111 / AUTOMATIC1111 Web UI)
Automatic1111 is the most popular Stable Diffusion interface. It stores generation parameters in PNG tEXt chunks with the key parameters. This chunk contains a structured text block with your complete generation recipe.
The tEXt chunk includes: full positive prompt, full negative prompt, steps count, sampler name (Euler, DPM++, etc.), CFG scale, seed value, image dimensions, model hash, model name, and any active extensions like ControlNet or ADetailer.
EXIF parameters
When saving as JPEG, Automatic1111 stores the same generation parameters in the EXIF UserComment field. Some interfaces also write to the EXIF ImageDescription and Software fields.
The EXIF data is structured identically to the PNG tEXt data. It contains the full prompt, negative prompt, and all generation settings in a parseable format.
ComfyUI workflow data
ComfyUI embeds the entire node workflow as JSON in PNG tEXt chunks. This includes not just the prompt and settings but the complete node graph, every connected node, and all parameters for each node.
ComfyUI workflow data is typically much larger than Automatic1111 parameters. A complex workflow can add 50-200KB of metadata to the image file.
Model hash and name
The model hash is a short identifier (like a1535d0a42) that identifies which checkpoint model was used. Combined with the model name, this reveals whether you used a base SD model, a fine-tuned model, or a community model from Civitai.
Model hashes can be looked up on Civitai and other model repositories to identify the exact checkpoint. This exposes your entire workflow.
Seed, steps, CFG scale, and sampler
These four values, combined with the prompt and model, are enough to exactly reproduce your image. The seed is a random number that determines the initial noise pattern. Steps controls iteration count. CFG scale controls how strictly the model follows the prompt. The sampler determines the denoising algorithm.
How StripShot removes Stable Diffusion metadata
- PNG files: Removes all tEXt, iTXt, and zTXt chunks. This eliminates Automatic1111 parameters, ComfyUI workflow JSON, and any custom metadata chunks.
- JPEG files: Strips APP1 markers containing EXIF data with generation parameters. Removes UserComment, ImageDescription, and Software fields.
- WebP files: Removes EXIF and XMP metadata segments that may contain SD generation data.
- All formats: Binary-level stripping. No canvas re-encoding. Zero quality loss. Original resolution preserved.
Processing happens entirely in your browser. Your images never leave your device.
Why removing SD metadata matters
Unlike DALL-E and Midjourney, Stable Diffusion does not use C2PA Content Credentials. Social platforms are less likely to auto-label SD images. But the generation parameters are a bigger problem for a different reason: prompt theft.
Anyone who downloads your image can open it in Automatic1111 (using the PNG Info tab) and see your exact prompt, negative prompt, model, seed, and settings. They can reproduce your image pixel-for-pixel. For artists selling AI art, this is a direct business risk.
SD metadata also reveals which model you used. If you trained a custom LoRA or used a specific Civitai checkpoint, the model hash and name are visible in the metadata. Stripping this protects your model choices and training investments.
Forge and other SD interfaces
Stable Diffusion Forge (a popular Automatic1111 fork) uses the same PNG tEXt metadata format. InvokeAI stores generation data in its own metadata format but also writes to PNG tEXt chunks. SD.Next and vladmandic forks follow similar patterns.
StripShot removes all PNG tEXt chunks regardless of the interface that created them. You do not need to know which SD interface was used. All generation metadata is stripped.
Related guides
Remove Midjourney Metadata
Strip Midjourney generation parameters and prompt data.
Remove Flux and Fooocus Metadata
Strip Flux AI generation parameters and PNG info chunks.
Remove DALL-E Metadata
Strip OpenAI EXIF tags and C2PA Content Credentials.
Remove All AI Metadata
Complete guide covering every major AI image generator.
Bypass AI Detection
Platform-by-platform guide for Instagram, Pinterest, Facebook.
Free. No sign-up required. Files never leave your device.