Loading...
Loading...
Numonic recovers generation data that AI tools embed in their files: prompts, settings and seeds, ComfyUI workflow graphs, models and LoRAs, and Content Credentials. How much comes back depends on the tool and the file type. This page lists it per source, including what is not read yet.
Checked against Numonic's extractors on .
| Source | Prompt | Settings & seed | Workflow graph | Models & LoRAs | Content Credentials | Key caveat |
|---|---|---|---|---|---|---|
| ComfyUIPNG | ReadOne positive, one negative | Partly readSeed, steps, CFG, sampler from the first KSampler node | ReadFull graph kept with the asset | ReadCheckpoints, LoRAs, VAE, ControlNet | Detected if present | Graphs built on custom samplers (SamplerCustom, common in Flux workflows) keep the graph but no settings. A seed fed in from a separate node is not read correctly yet. |
| ComfyUI videoMP4 | ReadOne positive, one negative | Partly readSeed, steps, CFG, sampler from the first KSampler node | ReadFull graph kept with the asset | ReadSame as ComfyUI PNG | Detected if present | Verified on MP4. ComfyUI WebM, MKV and AVI files import, but reading their workflow is not verified yet. |
| ComfyUI audioMP3 | Not readStays inside the stored graph | Not read | ReadGraph kept with the asset | Not read | Not in file | The graph is stored and the file is tagged as ComfyUI, but the prompt, settings and models inside it are not pulled out into searchable fields yet. |
| ComfyUI API nodesPNG, MP4 | ReadFrom the node’s prompt field | Not readNot read from API nodes | ReadFull graph kept with the asset | Partly readHosted model name, when the node has a model field | Detected if present | Verified with the Kling, Seedance, Nano Banana and GPT Image nodes. Other partner nodes are not verified yet. |
| ComfyUIJPEG, WebP | Not read | Not read | Not read | Not read | Detected if present | The workflow is read from PNG and MP4 only. Save as PNG if you want the graph recovered. |
| Midjourney, from midjourney.comPNG | ReadFull prompt text | ReadParameters such as --ar, --stylize, --chaos; seed only if you set --seed | Not in file | Not in fileMidjourney does not record models | Not in file | The Midjourney job ID is read from the file. Renaming the file does not matter; the data is inside it. A few downloads carry their details only in XMP, which is not read yet. |
| Midjourney, from Discord or re-savedPNG, WebP, JPEG | Not in fileUsually stripped from these copies | Not in fileUsually stripped from these copies | Not in file | Not in file | Not in file | A PNG copy that still carries the midjourney.com data is read like the row above. WebP and JPEG copies are not read. Download from midjourney.com to keep the prompt. |
| Automatic1111, Forge, SD.NextPNG | Partly readPositive; negative up to its first line break | ReadSteps, sampler, CFG, seed, size | Not in file | Partly readCheckpoint name and hash; LoRAs not read | Detected if present | PNG only. The same data saved inside JPEG or WebP is not read yet. |
| ChatGPT and GPT Image, DALL·E, Gemini and Imagen, Sora, Dreamina, FireflyPNG, JPEG, WebP | Not readNot written into the file | Not read | Not in file | Not read | ReadTool named for ChatGPT, Gemini and Dreamina images | These tools do not put the prompt in the file. Images made through the OpenAI API are flagged as AI-generated but not yet tied to a tool. Firefly exports whose credentials live in the cloud, not in the file, are not read. |
| InvokeAI, NovelAI, Fooocus, SwarmUI, Draw ThingsPNG | Not readNot recognized yet | Not read | Not read | Not read | Detected if present | The files import normally, but their generation data is not read yet. Exception: Fooocus images saved in the A1111 metadata scheme are read like Automatic1111 files. |
| Kling, Runway, Pika, Veo, Wan, LTX (downloaded directly)MP4 | Not read | Not read | Not in file | Not read | Detected if present | Stored as a video file with no generation data. Generate through ComfyUI to keep the graph, or add the prompt by hand. |
A recovered ComfyUI workflow tells you exactly what was used. It does not bring back the setup around it. Opening an old graph will not restore:
To keep a generation re-runnable, keep its models and inputs alongside it. The Full Archive export returns your files with their embedded metadata left in place.
Screenshots, format conversions, re-saves in an image editor and uploads to many social platforms remove embedded metadata. Numonic reads what is in the file, so data that is gone before import cannot be recovered.
Numonic records one positive and one negative prompt per file. In multi-pass ComfyUI graphs, the recorded prompt and settings come from the first prompt and sampler nodes found, not necessarily the final pass. The full graph is always kept.
A byte-for-byte copy of a file already in your workspace is recognized and not stored twice. A resized or re-saved copy counts as a new file.
Files you upload in the browser are checked for Content Credentials (C2PA); the check is on by default and can be switched off per upload. Files that arrive through Connected Folders are not checked yet.