Download and Load BLIP Model

This is the loader that feeds BLIPCaption. Its whole job is to get the BLIP weights onto disk and into memory, then hand out a BLIP_MODEL you plug into the captioner. The "Download And Load" name is the selling point - you don't have to hunt for the .pth file, drop it in the right folder, and pray the dropdown sees it. Pick a model name, run the graph, and it sorts the download for you.

If you've ever fought with "where do I put model_base_capfilt_large.pth" - and plenty of people have, it's one of the most-asked Art Venture questions - this node is the answer. Load once, wire its output into one or more BLIPCaption nodes, and the model stays resident instead of being reloaded on every caption. That's the pattern you want: one loader, reused.

How it works

Give it a model name and it ensures those weights are present (downloading them if they aren't), constructs the BLIP model object, and outputs it as a BLIP_MODEL. That output is an opaque handle - it doesn't render or preview anything, it just travels down a wire to a node that knows what to do with it. Here, that's BLIPCaption via its blip_model input.

The inputs and outputs that matter

• model_name - the BLIP checkpoint to load. Two choices:

  • model_base_caption_capfilt_large.pth - the captioning model. This is the one you want for turning images into descriptive sentences.
  • model_base_capfilt_large.pth - the base BLIP model. For plain image-to-text captioning, the caption variant above is the right default.

  If you're not sure, pick the caption one.

Output is a single BLIP_MODEL. Wire it into BLIPCaption's optional blip_model input. That's the entire interface - deliberately minimal.

How to install it

It's in the Art Venture pack. ComfyUI Manager → search comfyui-art-venture → Install → restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/sipherxyz/comfyui-art-venture

then pip install -r comfyui-art-venture/requirements.txt and restart.

Common issues & troubleshooting

The download fails or hangs. Serverless and locked-down environments sometimes can't reach the model host, or the run times out mid-download. The fallback is manual: grab the .pth yourself and place it in ComfyUI/models/BLIP - that's the folder the pack looks in, and it's the fix that keeps coming up in the community. Once the file's there, the loader uses it instead of downloading again.

It re-downloads every run. It shouldn't once the weights are cached, but if the model folder isn't persistent (a fresh container each run, no mounted cache), you'll re-fetch each time. On a platform with a persistent models volume this is a one-time cost; on ephemeral setups, pre-bake the file into your image or mount a cache.

BLIPCaption still says the model's missing. Make sure the wire is actually connected - DownloadAndLoadBlip's BLIP_MODEL output must go into BLIPCaption's blip_model input. If they're not linked, the captioner falls back to loading on its own and can hit the same folder problem.

Node missing after a workflow import. The pack isn't installed - Manager → Install Missing Custom Nodes pulls the whole set.