Wan 2.7 on VdoBloom: Text, Image and Reference Video Plus the Image Models, and When to Pick Each
Wan 2.7 on VdoBloom explained: the five variants, text-to-video prompts, first and last frame control, reference-to-video, and when Wan 3.0 wins instead.
Wan 2.7 is Alibaba’s current Wan generation on VdoBloom, and it ships as five separate tools: text-to-video, image-to-video, reference-to-video, Wan 2.7 Image and Wan 2.7 Image Pro. Video clips run 5, 10 or 15 seconds at 720p or 1080p, and a 5-second 720p render costs 32 credits while a 15-second 1080p render costs 143.
Updated October 2026.
What Wan 2.7 is and how it sits between Wan 2.6 and Wan 3.0
The Wan family is Alibaba’s open video and image line. On VdoBloom, 2.6 is the previous video generation, 2.7 is the current one, and Wan 3.0 is the longer-shot tier. The differences are practical rather than marketing labels.
Wan 2.7 keeps the 5, 10 and 15 second durations and the 720p and 1080p tiers of 2.6, and adds things 2.6 does not have: generated audio on the text-to-video model, first- and last-frame control on image-to-video, a reference-to-video variant, and two still-image models in the same generation. Wan 3.0 moves in a different direction, with shots that run to 30 seconds. If your clip fits inside 15 seconds and a still-to-motion pipeline, 2.7 is the newer and better-equipped choice. If the shot needs to run 20 to 30 seconds without a cut, that is a Wan 3.0 job.
The whole family carries a HOT content label on VdoBloom, the more permissive filtering tier, and the video variants are rated 4.3 by users. Platform-wide rules still apply to every generation. The tier above is covered in the Wan 3.0 guide, and the previous generation in the Wan 2.5 and 2.6 guide.
The five variants: text, image, reference video, Image and Image Pro
| Variant | Input | Output | Starting credits |
|---|---|---|---|
| Wan 2.7 Text-to-Video | Prompt only | 5/10/15s, 720p or 1080p, five ratios, with audio | 32 for 5s 720p |
| Wan 2.7 Image-to-Video | First frame, optionally last frame | 5/10/15s, 720p or 1080p | 32 for 5s 720p |
| Wan 2.7 Reference-to-Video | Reference images or video | 5 or 10s, 720p or 1080p | 32 for 5s 720p |
| Wan 2.7 Image | Prompt or edit instruction | 1K/2K/4K, eight ratios from 1:8 to 8:1 | Shown before you generate |
| Wan 2.7 Image Pro | Prompt or edit instruction | Same controls, 4K in text mode | Shown before you generate |
Text-to-video is the only variant that produces sound. Image-to-video and reference-to-video animate what you supply, and the reference model accepts a clip as well as stills. The two image models share their controls; Pro is the higher-rated tier at 4.4 and the one that delivers 4K in standard text mode.
Text-to-video settings and a worked prompt
The text model offers five aspect ratios: 16:9, 9:16, 1:1, 4:3 and 3:4. Pair the ratio with the placement first, because re-rendering for a different frame costs the same as the first render. Durations are 5, 10 or 15 seconds and resolution is 720p or 1080p. Audio arrives with the picture, so write the sound into the prompt instead of leaving it to chance.
A worked prompt for a 10-second 16:9 travel beat: “An empty two-lane road runs toward red rock mesas at first light; a single car passes left to right and its dust settles slowly behind it; camera locked off at chest height with a long lens; warm low sun, deep shadows; engine note fading, wind over dry grass.”
Three things make that prompt fit the model. The subject is named once, the action has a start and an end, and the sound is explicit. At 10 seconds the camera stays locked, which avoids asking a 10-second generation to do two jobs. The same prompt at 15 seconds should get a second beat, such as the car stopping and a door opening, or the model will stretch the single action and lose energy.
Reference-to-video for a character that stays the same
Reference-to-video is the variant to use when the face, product or art style has to survive across several separate generations. You supply reference images or a reference clip, and the prompt controls what happens rather than what the subject looks like.
Use it for a character sheet turned into a series of story beats, a brand mascot across a batch of 10-second ads, or a product that has to look identical in six different shots. The practical habit is to keep the reference set fixed and change only the prompt between runs. If the reference set changes, the result changes, and comparing the outputs tells you nothing.
Clip lengths are 5 or 10 seconds at 720p or 1080p, so a longer scene is planned as separate reference-locked shots rather than one long take. That is the same continuity problem covered in AI video character consistency, and a fixed reference set is the simplest answer to it.
Wan 2.7 Image and Image Pro for stills
The image side is unusually wide. Both models offer eight aspect ratios, including 1:1, 3:4, 4:3, 9:16, 16:9 and 21:9, plus the extreme 1:8 and 8:1 banner formats that few generators render natively. Both have a 1K, 2K and 4K ladder, both edit as well as generate, and both support an optional gallery mode.
The choice between them is narrow. Wan 2.7 Image reaches 4K, and Pro delivers 4K in standard text-to-image mode with the same controls. If you need extreme banner proportions or a 4K still straight from a prompt, Pro is the tier built for it. For screen-sized work, the base model is the cheaper place to draft. In both cases the credit cost appears before you generate, so there is no surprise at 4K.
Open image generation to draft at 1K and step up once the composition is right, or the image editor for instruction-based changes.
When to use Wan 3.0 or Seedance 2.5 instead
Pick Wan 3.0 when the shot has to run to 30 seconds in one generation. Wan 2.7 tops out at 15 seconds on every video variant, and stitching two 15-second clips introduces a join you then have to hide.
Pick Seedance 2.5 when you want a 4-to-30-second clip with per-second control and native audio, especially at 21:9, which Wan 2.7’s video models do not offer. The comparison is covered in Wan 3.0 vs Seedance 2.5 for the long-take question, and the same logic applies one tier down.
Stay with Wan 2.7 when you want first- and last-frame control at 720p or 1080p, a reference-locked character, or a still and a video from the same family in one sitting. It is also the cheaper option for a 5-second 720p test at 32 credits.
Frequently asked questions
How long are Wan 2.7 video clips?
Text-to-video and image-to-video offer 5, 10 or 15 seconds. Reference-to-video offers 5 or 10 seconds. All run at 720p or 1080p.
Which Wan 2.7 model generates sound?
Text-to-video, which produces audio alongside the picture. The image-driven variants animate your supplied frames or references.
What is the difference between Wan 2.7 Image and Image Pro?
They share controls, aspect ratios and the 1K/2K/4K ladder. Pro delivers 4K in standard text-to-image mode and carries a slightly higher user rating, 4.4 against 4.3.
How much does a Wan 2.7 clip cost?
A 5-second 720p render is 32 credits and a 15-second 1080p render is 143. Every combination is priced before you generate, and failed generations are refunded.
Does Wan 2.7 support first and last frames?
Yes, on image-to-video. Supply an opening image alone for an open-ended animate, or both endpoints to plan where the shot lands.
Open text to video for the audio version or image to video for frame control, pick Wan 2.7, and the credit cost appears before you generate. New accounts start with 10 free credits and no card; one-time credits never expire if you would rather not subscribe, and pack options are on the pricing page.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →