Recraft V4 Text to Image

This is where Recraft currently ships its best image model. The V3 family (the "red panda" leaderboard-topper) is still around as a workhorse, but V4 - and since July 2026, V4.1 - is the flagship: newer, bigger outputs, and a model dropdown instead of a style-object input. If you want Recraft's best pixels, this is the node.

The headline gotcha is right there in the schema, so let's get it out of the way: the negative prompt input is ignored. Not "softly ignored" - the official tooltip says verbatim that it's "not supported by Recraft V4 and V4.1 models." The input still exists in the UI because the schema needs it, and beginners will absolutely fill it in and wonder why it does nothing. Save yourself the confusion: put your "don'ts" into the positive prompt ("no text, no watermark, plain background") and leave the negative field alone. This is the single most common footgun on this node.

The inputs that matter

• prompt - multiline, up to 10,000 characters. Plenty of room.
• model - a dynamic combo. The main picks are recraftv4, recraftv4_1 (the July 2026 update), and the _pro variants (recraftv4_pro, recraftv4_1_pro). The pro models default to 2048x2048 output and cost notably more; the regular ones default to 1024x1024.
• n - 1 to 6 images
• seed - re-run control only; results stay nondeterministic regardless
• optional recraft_controls - color/background steering from the Recraft Controls node

Note what's not here: no recraft_style input, no negative prompt that works, no size combo at the top level. V4 dropped the V3 style-object system - the model dropdown and its per-model size option replace it. So the style nodes from the V3 family have nothing to plug into here; don't try to reuse them in the same branch.

How it works

Same hosted mechanism as the rest of the family: your prompt and settings go to /proxy/recraft/image_generation, Recraft generates on their GPU, the image is downloaded back into a tensor. Billed per image from your Comfy account credits. Ships with ComfyUI core; needs a Comfy account, no model files.

Common issues

• Pro vs regular is a real decision. The pro models are meaningfully more expensive per call and default to 2K output. If you're iterating on a concept, run the regular model and only spend pro credits on the finalist. There's no point paying for 2048x2048 while you're still deciding on composition.
• No style system means you style via prompt. V4's look is controlled by what you write, not by a style wire. If you loved the V3 substyle workflow, adjust your expectations - this node trades that for a stronger base model.
• Prompt length is a real resource here. With 10,000 characters available, verbose prompting is viable in a way it isn't on most models. Use the room for detail and constraints, because your negative prompt is a corpse.

Reach for this when you want Recraft's current best and don't need the V3 style ecosystem. Just remember the negative prompt does nothing, and decide upfront whether each run deserves the pro model's price tag.