Official ComfyUI docs and model cards give few VRAM, RAM or disk figures, and those that exist carry conditions such as offloading, precision or a specific runtime. This guide compiles them for SD1.5, SDXL, FLUX.1, Qwen-Image, Wan 2.2 and LTX-Video, marks gaps, and gives a procedure to record your own result.
Handing ComfyUI work to a client means keeping the assets, both workflow JSON formats, version and model identity, seeds and a licence record. Pixel-identical re-runs cannot be promised, so state in the delivery note what the package does and does not support.
ComfyUI's cited sources document listen addresses, TLS and Manager security levels but describe no built-in login, so a shared server needs authentication supplied by your own network layer. This guide covers access policy, custom-node risk, storage separation, backups and the Cloud trade-off, with limits stated.
Labels for people, the workflow ComfyUI embeds in files and signed C2PA Content Credentials are three different things. This guide summarises the cited sources, including EU AI Act Article 50 as of 2026-10-06, and a recording procedure; it is not legal advice.
Licences differ sharply between model families, and a LoRA or finetune can add its own terms. This guide gives a record-keeping checklist and summarises what the licence texts for SDXL, FLUX.1, Qwen-Image (and Qwen-Image-2.1), Wan 2.2 and ComfyUI state, without legal advice.
The U.S. Copyright Office says protection covers human authorship, requires disclosure of more-than-minimal AI-generated material, and treats prompts alone as insufficient control. This guide turns that into a record-keeping routine for illustrators and publishers, and states what such a record cannot prove; it is not legal advice.
Teams using ComfyUI and Wan models for previsualization and animatics need shot plans that fit what the model cards state, a clear handoff format for editorial, and per-clip provenance. This guide sets out an editorial procedure from the cited official pages.
Editing nodes can recolour or restyle garments, but the cited pages do not promise that a product is depicted accurately. This guide covers checking product fidelity, consent for real models and disclosure questions under the EU AI Act, New York model-consent text and FTC guidance, as editorial research and not legal advice.
Concept visuals from sketches, plans or 3D screenshots can use ComfyUI control images and masked edits, but the results are illustrative and not dimensionally accurate. This guide covers choosing control images, checking outputs against the source drawing, labelling images for clients, and recording licence and provenance.
Advertising teams generating many variants through the ComfyUI API need a per-variant record linking parameters, prompt_id, output file and approver. This guide sets out that structure and points to the FTC, Google and Meta pages (read 2026-10-06) relevant to AI-generated or edited ad imagery; it is not legal advice.
Generated game art needs a per-asset record covering model identity and licence, version, transparency handling, human edits and store disclosure. This guide cites Steamworks, Unity Asset Store and model licence texts and gives an editorial checklist, not legal advice.
Online-store product images can have their backgrounds replaced in ComfyUI, but the product must remain faithful to the physical item. This guide covers masking decisions, batch submission through the API, marketplace requirement checks and the accuracy risk of altered images.
The last line of a screenshot rarely identifies the instance, input and first exception. Preserve a complete local record and share a deliberate redacted report.
A model name can cover bundled checkpoints, split weights, quantized variants, text encoders and VAEs. Preserve enough identity information to know which artifact a workflow actually used.
Supported extra_model_paths configuration can reference existing model storage. It does not share Python environments, node registration or workflow versions.
Identify the format and source of JSON, images or videos before installing nodes and weights. Static inspection helps planning but cannot replace execution in the target environment.
Recovering old code does not automatically restore Python packages, frontend, nodes or configuration. Establish recovery information for the whole relevant environment.
Replace random uninstalling with controlled groups and repeatable input. The goal is the smallest triggering set, not a blanket verdict on all third-party nodes.
Downloading a repository is only one step. Identify the owning package, satisfy its dependencies and confirm that the running backend actually registers the required class.
Portable contains its own Python and ComfyUI directories. Installing into system Python does not automatically make those packages available to its embedded interpreter.
Plan around the supported interpreter, device backend and a separate working directory, not a single pip command for every machine. Desktop and Portable users should follow their own installation paths instead.