PixVerse V4

PixVerse V4 is where the PixVerse text-to-video family in the Runware pack starts to feel like you're directing rather than prompting. Same shape as V3.5 - prompt in, VIDEO out - but V4 adds a proper cameraMovement dropdown, and that's the feature that matters. A crane up, a smooth zoom-in, a horizontal dolly: instead of hoping the model reads "slow dolly left" from your prompt, you pick the move and it happens. That's the difference between "video that exists" and "video with intent."

The rest of the node is the familiar PixVerse kit: frame images for image-to-video, size presets, duration, style, effects, motion mode. V4 is the version you reach for when the camera work is part of the brief and you don't want to cross your fingers.

What you actually set

• positivePrompt - required. The scene and the action; the camera move has its own input, so you can leave direction out of the prose.
• providerSettings.pixverse.cameraMovement - the headline feature: horizontal_left, horizontal_right, vertical_up, vertical_down, zoom_in, zoom_out, crane_up, auto, and a dozen more. Choose the shot.
• frameImages - IMAGE socket for image-to-video.
• size - 360p to 1080p in the usual aspect ratios, default 720p (16:9).
• duration - 5 or 8 seconds.
• providerSettings.pixverse.style - anime, 3d_animation, clay, comic, cyberpunk.
• providerSettings.pixverse.effect - the viral movement templates, shared with V3.5.
• providerSettings.pixverse.motionMode - normal or fast.
• negativePrompt, seed, numberResults - the usuals.

Output: one video socket, type VIDEO.

Install

Same one-time pack setup: ComfyUI 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

API key into ComfyUI Settings → Runware API key or RUNWARE_API_KEY. Nothing heavy installs; the models are cloud-side.

Where people get burned

• Camera movement fights the prompt. If you pick zoom_in and also write "the camera pulls back," you've given the model a contradiction and it will resolve it badly. Pick the move, describe only the scene.
• It's a paid upgrade over V3.5. V4 costs more per run and the price shows on the title bar. Only pay for it when the shot genuinely needs the direction.
• safety.mode defaults to none on the video nodes - content checking is opt-in.
• Effects and camera movement don't compose. Applying a meme effect plus a crane shot produces chaos. Choose one flourish per clip.

The verdict: V4 is the one I'd actually reach for out of the PixVerse line for anything that isn't a social-media loop. The camera dropdown removes the single most frustrating part of text-to-video - telling the model how to frame the shot.