Wan 2.7 Text to Video

Pure text-to-video on the newest Wan model, as a built-in node. You type a prompt, Wan 2.7 generates a clip, and - this is the interesting extra - you can feed it an audio track to drive the generation, or let the model cook up matching background music and sound effects on its own. It's the most "prompt in, movie out" node in the Wan family, and it's also the most honest about what the hosted Wan line is: Alibaba stopped releasing open weights after Wan 2.2, so 2.7 only exists through APIs, and this node is how you reach it from inside ComfyUI.

It's an API node like the rest of the built-in partner/video category: ships with ComfyUI core, sends your prompt (and optional audio) to Alibaba's Wan API through Comfy Org's backend, and bills your Comfy account in credits. Nothing runs on your GPU. The SEO-spam "download Wan 2.7 locally" pages you'll find in search results are a trap - there are no weights; this node is the real path.

The inputs that matter

The model input is a dynamic combo - a dropdown that unfolds the controls when selected. Inside wan2.7-t2v you get:

• prompt - the video description, in English or Chinese. This is 100% of the creative work; Wan 2.7 follows descriptive, concrete language well.
• negative_prompt - supported, so use it for what to avoid.
• resolution - 720P or 1080P. ratio - 16:9, 9:16, 1:1, 4:3, 3:4. duration - 2 to 15 seconds.

Outside the combo: audio (optional, 3–30 seconds - "for driving video generation, e.g., lip sync, beat-matched motion"; if you don't provide one, the model auto-generates matching BGM or sound effects), plus seed, prompt_extend (server-side enhancement, on by default), and watermark.

Output is a single VIDEO.

The honest take

Set expectations before you queue a render: Wan 2.7's community reception was muted to hostile. The biggest reaction thread ran 67 points of "it's shit tbh, compared to other closed source models," with people claiming plain Wan 2.2 looks better. The one substantive paid comparison found the truth is more nuanced - cleanest small-detail textures of the four models tested on close-ups (faces, fabric, water), but visible pose discontinuity on heavy character motion. So: static scenes, slow pans, and detail-heavy close-ups are where 2.7 earns its keep; big action is where it wobbles.

Where people get burned: expecting open weights (there are none - see the download-link farm warning above), expecting a free tier (per-second credits, and the badge is honest about it), and skipping the audio input on a model that genuinely shines at beat-matched or speech-driven generation. And if all you need is quality-at-any-price, remember the rest of this family - Veo and Kling - are also a dropdown away.