Generate Image (Venice)

Your ComfyUI workflow is likely running on whatever your GPU can hold - and if that's a 6 GB card, a full Flux checkpoint is not happening locally. Generate Image (Venice) swaps the local checkpoint for venice.ai's servers: you type a prompt, and Venice generates the image on its side and hands back a normal IMAGE tensor. Same node-graph feel, none of the VRAM. If you're already paying for Venice or just want uncensored-capable hosted generation without spinning up your own rig, this is the node.

Venice is worth knowing as a service, too: it's positioned as the "uncensored" hosted AI platform, and the community debates how well that holds up at higher tiers and for adult content. The node's own fields reflect that posture - there's a flux-dev-uncensored model, a safe_mode that blurs adult-classified output, and a hide_watermark toggle the author warns Venice can ignore for NSFW content.

How it works

The node POSTs your parameters to https://api.venice.ai/api/v1/image/generate, gets back base64 PNG data, and decodes it into a [B, H, W, C] tensor that plays nice with every preview and save node you already use. One call per image in the batch, then the tensors are merged.

The inputs that matter

• model - auto-populated dropdown, default flux-dev. The tooltip is the honest tell: if it only shows flux-dev or C O M B O, the model list fetch failed - usually because the API key is missing or wrong. Fix the key, restart, done.
• prompt - 1 to 1500 characters. Empty or over-length and the node refuses before it even calls the API.
• neg_prompt - ignored entirely for flux-dev and flux-dev-uncensored. That's not a bug: Flux is guidance-distilled, so it doesn't use negative prompts the way SDXL does. Fill it in for the SD/SDXL-family models, leave it for Flux.
• width / height - must be multiples of 32. The node lets you dial up to 2048, but Venice's real ceiling was 1280 at the time of writing - the API will error if you exceed it.
• steps - 1–50, and per-model caps apply. flux-dev and SD 3.5 max out at 30; most others allow 50.
• guidance - default 3.0, range 0–20. For Flux keep it in the low single digits (roughly 3–4); it's guidance scale, not SDXL-style CFG.
• style_preset - a Venice style gallery; none skips the field entirely.
• hide_watermark (default true) and safe_mode (default false) - the content-policy dials described above.
• seed - optional, default -1 (random). Batches step the seed per image, so one seed gives you a reproducible set.

Output

IMAGE - a standard ComfyUI tensor, ready to plug into a preview or Save Image node. From there it behaves like any locally-generated image, so downstream nodes don't care where it came from.

Installing it

One of seven nodes in the ComfyUI-Venice-API pack - install once, use them all. ComfyUI Manager (search "ComfyUI-Venice-API"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/DraconicDragon/ComfyUI-Venice-API.git
pip install -r ComfyUI-Venice-API/requirements.txt

Restart, then add your key in Settings > VeniceAI > API Key. No checkpoints to download, no LoRAs to stash - the only thing between you and a Flux image is a key and a network connection. Keep in mind the pack is a one-person passion project (a rewrite of the Together AI nodes), so treat version bumps as potentially breaking.