zhenzhen-grok-video-1.5-new

The budget lane of the Grok video family. Comfly_grok_video_1_5 calls the grok-1.5-video model - the display name says "new," and the changelog confirms it: 720P default, three lengths (6s / 10s / 15s), priced at 0.5 credits for the 6-second and 0.7 for the 10/15-second runs. It's image-to-video, so you feed it a starting frame and it animates from there.

Where this sits in the lineup: the grok-video-3 nodes are the flagship (multi-ref, 1080P, the Sora-stand-in). This is the older, cheaper sibling - fewer bells, fewer credits. Reach for it when you're iterating on motion ideas or need a quick i2v test without burning flagship rates.

The inputs that matter

• prompt - the motion instruction. Defaults to a Chinese template ("基于这张图片生成视频" - "generate a video based on this image"), so replace it with what you actually want the image to do.
• model - grok-1.5-video-6s, grok-1.5-video-10s, or grok-1.5-video-15s. Pick your length here; that's also your price tier.
• size - 1280x720 or 720x1280. Landscape or portrait.
• image - the starting frame (IMAGE tensor), or image_url if your reference is already hosted. The tooltip: image_url is ignored when the image input is connected.
• poll_interval / max_poll_attempts - the wait loop; the default 600 × 6s is a full hour, plenty.
• seed, apikey, skip_error.

Outputs: video (VIDEO tensor), response, video_url, task_id.

Install and key

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart, or ComfyUI Manager → "Comfyui-zhenzhen". Key from ai.t8star.org into apikey - the README puts this one on the default token group.

The honest version

You're trading quality and resolution for price here, and that's the deal - don't ask a 0.5-credit 720P clip to look like a flagship render. Feed it a clean, well-composed starting frame, because i2v quality is mostly downstream of the frame you give it. Standard pack rules hold: transient 500s re-run clean, task_id gets you into the site's async list if ComfyUI times out, and the model's content filtering applies. For fast motion roughs, it's one of the cheapest ways to get a video out of this pack.