HappyHorse-图片文字生视频

The sibling of the pack's HappyHorse reference-video node, and the more flexible of the two. This one takes a text prompt and optionally an input_image - so you can go pure text-to-video when you have no still, or drop an image in when you want the scene to start from something real. Same model family, same strengths and the same honest caveats: steady on objects, shaky on complex motion, and notably relaxed content review compared to the big US video models. For e-commerce b-roll that's a genuinely good trade.

How it works

The prompt (plus optional input_image) goes to the gateway under the model id HappyHorseTI2vNode. No image wired? The model invents the opening frame from your text. Image wired? It uses it as the starting point and animates from there - effectively turning this node into a hybrid between the reference node and a pure T2V. The clip returns as a VIDEO tensor at your chosen resolution and duration. The README's tips carry over: it's a product-stability model first and a motion model second, so write prompts around gentle, saleable motion.

The inputs that matter

• prompt (STRING) - the whole scene. This is your only "image source" when input_image is empty, so describe the composition, the lighting, and the small amount of motion you want.
• input_image (IMAGE, optional) - a starting still. Wire it in and you're doing image-assisted T2V.
• resolution (720P / 1080P) - output size.
• duration (INT, 3–15, default 5) - seconds.
• Size (1:1, 3:4, 4:3, 9:16, 16:9, 5:4) - aspect ratio.
• seed (INT) - reproducibility.

One output, video (VIDEO).

Installing it

Same pack, same drill - comfyui-MJAPI-party. ComfyUI Manager → search "mojieapi_party" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MoJIeAIGC/comfyui-MJAPI-party.git

Register at mojieaigc.com, add your key to config.ini (leave BASE_URL = https://www.mojieaigc.com/v1/completions alone), restart:

[API]
KEY = your_api_key
BASE_URL = https://www.mojieaigc.com/v1/completions

No model downloads.

Common issues

• Without an image, the first frame is a guess. Pure T2V means the opening composition is the model's invention. If the start matters, wire in an input_image - that's the entire difference between this node and a frustrating dice roll.
• Motion remains the weak spot. Same as the reference node: keep the action simple and the subject stable; that's what this model is actually good at.
• Name confusion, same as the sibling. Don't go hunting for HappyHorse "open source" weights - the community has flagged fake repos under that name. Here it's an API call through the gateway; nothing to download, nothing to vet.
• Error image with text - the pack's standard failure display (key, balance, timeout).