Model · Alibaba
Wan, open-weight generative video
Wan is Alibaba's video family, from an open-weight lineage — which is why it turns up everywhere in the ecosystem. Four engines are wired in here, from fastest to most capable, with no GPU to rent and nothing to install.
What Wan brings
Long shots
Wan 3.0 and 3.0 Prime reach 30 seconds where most video models stop at 10 or 15. For a product demo or an atmosphere sequence, that avoids stitching three shots that do not match.
Ten images and five reference clips
Wan 3.0 accepts up to ten input images and five reference clips. So you impose a subject, a set and a movement instead of describing them and hoping.
An adaptive aspect ratio
The 3.0 releases handle adaptive framing alongside fixed 16:9, 4:3, 1:1, 3:4 and 9:16 — useful when the reference does not share the output format.
A reference audio track
All four engines accept a reference audio track, so the pace of the shot can be set against existing music or a voice rather than the other way round.
The four Wan engines available here
From fastest to most capable, in the same picker.
- Wan 3.0 PrimeText, image or reference to video. 2–30s, 480p to 1080p, up to 10 images and 5 reference clips.
- Wan 3.0Same inputs and durations, the standard tier of the current generation.
- Wan 2.7Text or image to video, 2–15s, 720p or 1080p, up to 2 input images, five ratios.
- Wan 2.6 FlashFast image-to-video, one input image required, 5–15s, 720p or 1080p.
What Beemm Vision adds
Wan also runs locally if you own a GPU. The point here is different: the chain.
- Nothing to install, no weights to download, no GPU to rent — and a graph that worked last month still works.
- The image generated at the previous step wires straight into the video node.
- Compare Wan, Seedance, Kling and MiniMax on one brief before committing to a 30-second generation.
- Derive the video into several formats without redoing the settings one by one.
- One credit balance, and closed models sitting next to open ones in the same graph.
Frequently asked questions
- How long can a Wan video be?
- Wan 3.0 and 3.0 Prime go up to 30 seconds. Wan 2.7 stops at 15 seconds, as does Wan 2.6 Flash. The picker shows the exact range for each before you launch.
- Do I need a start image?
- It depends on the engine. Wan 3.0, 3.0 Prime and 2.7 accept text alone; Wan 2.6 Flash requires an input image. The picker says which.
- Wan is open-weight — why go through you?
- To avoid installing, maintaining or renting a GPU, and so Wan can sit in the same graph as closed models — GPT Image, Gemini, several video models — that are not weights you can download.
- Can I supply reference music?
- Yes, all four engines accept a reference audio track, so the pace of the shot can be set against an existing piece or a voice.
Generate a video with Wan
Start from text, from an image, or from references you have just generated on the same surface.
Keep reading
- GPT Image on Beemm VisionUse OpenAI's GPT Image 2 and GPT Image 1.5 without a ChatGPT subscription: generation, editing and eight formats, on a balance shared with the whole catalogue.
- Hunyuan Image on Beemm VisionGenerate with Tencent's Hunyuan Image: text-to-image and editing, output up to 4K, eight formats — in a canvas that gathers the whole catalogue.
- Nano Banana on Beemm VisionUse Google's Nano Banana 2 and Nano Banana Pro image models on one canvas: legible in-image text, conversational editing, output up to 4K.
- The full catalogueEvery image and video model available in the canvas.