BYOKey MiniMax Image to Video

MiniMax - the company behind Hailuo - is one of the stronger hosted video generators, and BYOKey_MiniMax_Image2Video is how it joins your ComfyUI graph. It takes an image as the first frame, adds a text prompt, and returns a real ComfyUI VIDEO, calling api.minimaxi.chat/v1 directly with your key as a Bearer token. No Comfy proxy, no credits.

The model list is the interesting part here: I2V-01 is the standard image-to-video model, I2V-01-Director is the one with camera/direction control, and I2V-01-live is the faster, more stylized live-action variant. Same image-to-video base, three different behaviors - that's a degree of choice most of the pack's video nodes don't give you.

How it works

Submit-and-poll like the other video nodes: POST the generation, poll until completion, download the result into a VIDEO output. One implementation detail worth knowing: because this is a direct call rather than going through Comfy's proxy, your first-frame image is sent as an inline base64 data URI - the README notes MiniMax's API accepts that format natively, so no upload step is needed. Big images mean big requests, but it works.

The inputs that matter

• image - the first frame of your video. This is the input that makes it image-to-video; without it you're in the wrong node.
• prompt_text - the prompt guiding the motion, up to 2000 characters. Describe the movement and the scene evolution, not the static contents.
• model - I2V-01 (default), I2V-01-Director, or I2V-01-live. Director is the one to reach for when you care about the camera move; live for faster, more casual results.
• prompt_optimizer - a boolean, default true. MiniMax's automatic prompt expansion. It's on by default for a reason: MiniMax's results improve noticeably when the prompt is enriched. Turn it off only if you need your prompt passed through untouched (or if you're getting results that drift from your intent).
• seed - set it if you like; honest tooltip territory, don't bank on reproducibility.
• poll_interval (default 10s, min 3) / max_wait (default 600s, up to 3600) - the queue knobs.

Output is video (VIDEO) - wire it into a video preview or save node to see it.

Installing it

It's one of 37 nodes in comfyui-byokey, with zero extra dependencies beyond ComfyUI's bundled torch/aiohttp/PIL/numpy/av. Install via ComfyUI Manager (search "byokey") or:

cd ComfyUI/custom_nodes
git clone https://github.com/MeteorAndy/comfyui-byokey.git

Restart, and it's under api/byokey/video. No models download - MiniMax renders it.

Gotchas

The recurring video-node cautions apply: cost (video is the expensive end of the pack, and Director/live tiers differ), the synchronous wait (raise max_wait if a busy queue times you out), and the source-image quality floor - a soft first frame gives soft motion, so prep the still before feeding it in. One MiniMax-specific thing to watch: prompt_optimizer on can pull results toward the kind of scene MiniMax "thinks" you meant, so if the output keeps drifting from your specific intent, flip it off and write a more detailed prompt instead of fighting the optimizer. Standard pack hygiene: your key is saved into the workflow JSON, so scrub it before sharing. For animating a still with Hailuo's motion quality, this is the node.