Runway Gen-4 Image Turbo

Here's the thing that makes Runway's Gen-4 line different from almost every other text-to-image model you've used: it's reference-first. The referenceImages input isn't optional style guidance - it's the required core of the model. You give Gen-4 one or more images, and it generates new images that hold the identity, the character, the objects from those references. That's the "reliability" Runway has been selling since Gen-3: consistent characters and subjects across generations, without training a LoRA.

This is the Turbo version of the image model, which means the fast distilled lane. Reach for it when you're iterating on characters - "keep this person, this costume, this object, put them in a new scene" - and you want quick turnaround. The same identity-preservation trick local workflows spend an IP-Adapter node and a lot of tuning to get is the default behavior here.

How it works

referenceImages (IMAGE, required) + positivePrompt go to runway:4@2 with taskType: imageInference; the result returns as a native image (IMAGE) tensor. Note the reference images travel as tensors - unlike the video nodes' UUID/URL inputs, this one wires directly from Load Image or any IMAGE-producing node.

Inputs that matter:

• referenceImages - required. The character/subject you want preserved.
• positivePrompt - the new scene, pose, or style around the reference.
• width / height - 1024×1024 default.
• numberResults - up to 20 seeded variations.
• providerSettings.runway.contentModeration.publicFigureThreshold - auto/low; loosen if recognizable faces are getting blocked.
• outputFormat / outputQuality - JPG/PNG/WEBP.

Output: image (IMAGE).

Installing

Pack-wide: ComfyUI Manager → Custom Nodes Manager → search Runware → install → restart, or:

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

No model downloads, light deps (runware-sdk, Pillow). API key from the Runware dashboard.

Gotchas

The reference is the contract - a blurry, low-contrast, or inconsistent reference image produces exactly that character in every output. Give Gen-4 a clean, well-lit shot and it does its best work. Second, "Turbo" means fewer iterations and lower cost per image, but also a quality cap versus the full Gen-4 pipeline; use it for drafts and character exploration. And remember the content-moderation dial: if you're getting refusals on a reference that looks like anyone public, publicFigureThreshold: low is the first thing to try before you rewrite your whole prompt.