Load FlashVSR Prompt

If you've loaded a FlashVSR workflow and wondered where the text prompt went - there isn't one. FlashVSR, the one-step 4x video super-resolution model from OpenImagingLab, conditions on a fixed prompt tensor that ships with the checkpoint, not on text you type. This node loads that tensor and hands it to ComfyUI as CONDITIONING, so the rest of your graph treats it like any other conditioning. You don't write a prompt here; you just pick the file and wire it to BasicGuider.

This is one of those details that makes sense once you think about it. FlashVSR is a restoration model, not a generation model. Its conditioning exists to give the DiT a stable, model-consistent "task prompt" - upscale this video - and the released Prompt.safetensors encodes that. There's nothing for you to author, and if you're reaching for a CLIPTextEncode node to give it a text prompt, stop: the loader's output is already the right conditioning for the graph.

How it works

It scans ComfyUI/models/flashvsr/ for a file with "prompt" in the name (the released Prompt.safetensors), reads the stored context tensor, and returns it as a CONDITIONING - specifically a [[context_tensor, {}]] batch, the exact shape ComfyUI's sampling shell expects. One input (prompt_name dropdown), one output (CONDITIONING).

In the intended graph, that output feeds BasicGuider alongside the patched model from Configure FlashVSR Upscaling, and the FlashVSR One-Step Sampler does the rest. The faithful sampling modes need the single conditional pass that BasicGuider provides, so keeping the loader's output on that path matters - don't reroute it through some other conditioning assembly.

Installing it

Part of the ComfyUI-FlashVSR-Stock pack - one install, ten nodes. ComfyUI Manager: search ComfyUI FlashVSR Stock Wan, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Pizzawookiee/ComfyUI-FlashVSR-Stock.git
cd ComfyUI-FlashVSR-Stock
python -m pip install -r requirements.txt

Restart, then download Prompt.safetensors from pizzawookiee/FlashVSR-1.1 on Hugging Face into lowercase ComfyUI/models/flashvsr/. Python deps are light - einops, safetensors, plus a Triton runtime.

Common issues

• "No prompt safetensors found in models/flashvsr". The file isn't in the lowercase flashvsr model folder, or you haven't restarted since placing it. It's the same three-cause answer as every loader in this pack.
• The workflow errors about conditioning shape. You probably connected the loader's output somewhere unusual, or let a different conditioning node feed BasicGuider. Wire this node's CONDITIONING directly into the guider.
• You're tempted to "improve" the prompt. Don't. It's a fixed checkpoint component; the whole point is that it stays stable across runs so FlashVSR's one-step path behaves reproducibly.

The node's job is to be boring: pick the file, wire one output, move on. The moment you're fiddling with it is the moment you're fighting the model's design.