Wan 2.2 speech is not a talking head

Wan 2.2 Speech to Video is not a talking-head generate. It will not cast a host, build a studio, or hear a script typed into a prompt. It is a lipsync row: static image plus audio in, talking picture out, 720p ceiling, 50 credits listed. If you wanted a presenter invented from a paragraph, you wanted a different Wan — or a different family.

The meter for plate-plus-track is already on the site. This page is the mix-up that wastes the 50-credit floor: treating speech-to-video as text-to-video with a mouth.

Lipsync categories, not text-to-video

The published slug is wan-2-2-14b-speech-to-video-lipsync. Content type is lipsync. Categories are audio-to-lipsync and image-to-lipsync. An image is required. Audio is on. Max output is 720p, with 480p, 580p, and 720p listed. There is no duration ladder; length follows the track. The description adds facial expressions, body movement, and camera work on top of the mouth — still driven by the two files you brought, not by a scene prompt.

That combination fools people. “Body movement and camera work” sounds like a commercial generator. It is not. The room is the still. The words are the wav. Extra motion is performance on that plate: shoulders, a push, a bit of life. It is not a location scout.

Wan 2.7 Text to Video is the family member that starts from a sentence: 2s–15s in one-second steps, 1080p ceiling, 15 credits, optional custom audio, prompt rewriting. That row can invent a shot. It is not this slug. Sending a “talking founder in a loft” paragraph to 2.2 Speech is how you discover requires_image the hard way — or worse, how you upload a random still so the job will start, then watch the wrong person say the right line.

Run 2.2 Speech from AI lipsync, not from a text-to-video box. The AI video generator is for rows that invent shots. Mixing launchers is how the brief loses its second file.

720p talking is a close-up spec

The ceiling is 720p. There is no 1080p or 4K on this output list. That is a talking close-up spec, not a cinematic master spec. Crop for the face. Stop apologising. If the master must be 4K talking, this is the wrong Wan. Kling Avatar Pro lists 4K on a lipsync card. Different family, different floor (58 credits), same two-file logic. Do not upscale a 720p mouth and call it a 4K presenter.

Body movement in the description is a crop decision. A tight beauty still cannot grow shoulders. A full-body fashion plate will not sit like a newsreader. Give the still the body you want to see move. Then do not ask the model to invent a second character or a product that is not in the pixels.

Camera work is the same rule. A small push on a locked plate is in-family. A new location is not. If you need the camera to travel between two approved compositions, that is first-last frame on a generate row, not speech-to-video.

Fifty credits is the listed figure and the matrix floor. Billing is duration-driven. A two-minute monologue is a two-minute bill. Cut the audio to the line you mean. A “cheap talking test” that includes the ums is how this SKU becomes the most expensive still you ever uploaded.

What Wan 2.7 is for instead

Use Wan 2.7 when the job is a shot, not a mouth.

• A scene that does not exist yet: loft, kitchen, street, product on a table with camera motion you will describe.
• Identity hold from a still without a speech file: Wan 2.7 Image to Video is 2s–15s, 1080p, 15 credits, first-frame / first-and-last / continuation. Catalog audio is unset. Optional driving audio is a track you attach, not a soundtrack the model always invents. That is still not lipsync.
• A 1080p plate that will never show a talking jaw.

Do not use 2.7 as a talking-head shortcut because 15 credits looks cheaper than 50. A 2.7 shot that happens to contain a mouth will not viseme your script. You will caption a closed mouth or burn a line that does not match. The cheap row is cheap at being a shot.

If you have no still and you need a talking person, you still do not start on 2.2 Speech. Generate or shoot the still. Generate or record the wav. Then lipsync. Premade avatars are a different catalog family. 2.2 Speech is not a casting library.

Native-audio cinematic talking — always-on dialogue in the generate — is a different job again. Native audio is Veo-shaped scene clips, not a Wan 2.2 mouth pass. Do not import that expectation onto a 720p lipsync SKU.

Two files or you are in the wrong office

Hold the still. Play the audio with your eyes closed. If either file would embarrass you, do not combine them.

If you have a sentence and no files, you wanted text-to-video or a stills row plus TTS, in that order, not speech-to-video.

If you have a still and no audio, you wanted image-to-video or a silent plate. 2.2 Speech will not read the prompt aloud.

If you have audio and no still, you wanted a premade avatar or a talking generate that does not require an image. This slug requires an image.

Captions still go on afterwards. A moving mouth is not a transcript. Add captions once the take is the take.

Article 50 still applies if the talking picture resembles a real person and you publish it as authentic-looking media in the EU. Article 50 disclosure is the label copy. A 720p viseme pass is not an exemption.

The 50-credit, 720p, plate-plus-track recitation lives on Wan 2.2 Speech to Video: the mouth pass after the picture exists. This page is the category error: speech-to-video is lipsync. It is not a talking-head generate.

If both files are good and the delivery is a 720p talking close-up, sign up and run the lipsync row on purpose.

FAQ

Can Wan 2.2 Speech to Video start from text only?

No. The catalog requires an image. Categories are audio-to-lipsync and image-to-lipsync. A prompt without a plate is a different Wan.

Is this the cheap way to make a talking presenter?

It is a 50-credit, 720p mouth pass on files you already approved. Inventing the presenter is stills plus speech, billed separately. Skipping those steps does not make 2.2 cheaper. It makes it a miss.

Should I use Wan 2.7 if I also want the mouth to move?

Only if you do not have a line to match. If you have a script as audio, 2.2 Speech is the mouth job. If you have a scene to invent and no viseme contract, 2.7 is the shot job. Do not ask one row to be both.