A car commercial ends with the vehicle parked on a coastal road. The director wants five more seconds: the camera slowly pulls back, the sun drops behind the cliff, and the taillights glow. There is no budget for another shoot day. Wan 2.7 takes the last frame of the existing clip, extends the scene, and keeps the car, the road, and the lighting consistent. The continuation cuts into the original without a visible seam.
Wan 2.7 is the latest video model available in OpenVideoMaker. It handles text-to-video, image-to-video, first-frame and last-frame control, reference video workflows, and scene continuation with flexible duration and prompt extension. The model works best when you give it a clear subject, a defined motion idea, and a reference that anchors the visual direction. It struggles when the prompt asks for too many competing actions or when the reference clip is too long or too busy.
OpenVideoMaker is a multi-model workspace, which means you can move between models without losing context. You might generate a product still with Seedream, feed it into Wan 2.7 for a reveal animation, compare the result against a Seedance pass, and then reuse the strongest clip as a reference for the next iteration. Each model rewards a slightly different brief, so the workflow matters more than any single prompt.
Getting started with Wan 2.7
Before you open the generator, answer four questions: what asset are you making, who will see it, where will it appear, and why does it need to exist? A product reveal for a paid social test, a scene continuation for a longer edit, a reference-guided motion test for a pre-production concept, and a social teaser for a launch campaign all need different instructions. If you skip this step, the model will still produce something, but you will have no clear way to judge whether the result is useful.
For Wan 2.7, the inputs that matter most are text prompt, first frame, last frame, image references, video reference, duration, and output quality. Treat each one as a control surface. The prompt controls language and intent. The reference image controls subject and composition. The first and last frame control the start and end state of the motion. The duration controls pacing. The model choice controls the tradeoff between speed, polish, reference handling, and motion behavior. The review checklist controls whether the team keeps the result or regenerates.
Do not start with a long prompt. Start with a compact brief that names the subject, setting, desired change, camera style, and output purpose. Then expand only when the output shows a specific weakness. If the product is drifting, add product-specific traits. If the scene is too static, add motion language. If the video is visually busy, remove secondary actions and keep one main motion idea.
How Wan 2.7 fits into a production workflow
OpenVideoMaker works best as a connected workflow, not a one-off generator. A typical path starts with a content goal, moves into image or video creation, then loops through prompt refinement and asset review. Each generated asset should become more useful in the next step, not simply add clutter to the asset library.
For image-heavy projects, start by generating or selecting a clean reference frame. Use GPT Image, Seedream, Imagen when the project needs still images, product concepts, references, or visual direction. Once the still frame is working, continue into Seedance, Veo, Kling if the campaign needs motion. For video-heavy projects, begin with the motion brief, then decide whether a source image, first frame, last frame, or reference video would give the model a better anchor.
Users looking for a direct workflow should enter through Wan Video Generator. Users comparing broader options should browse AI Video Generator or AI Image Generator. Users who need prompt help should review Image to Video Prompts, Product Video Prompts, or Seedance Prompts.
What Wan 2.7 does well
- Flexible duration: generate clips at different lengths depending on the channel. A 3-second product reveal for a social story, a 5-second continuation for a longer edit, or a 10-second atmospheric clip for a brand video. Set the duration to match the output need instead of trimming later.
- Prompt extension: the model can extend a scene from an existing clip, keeping the subject and environment consistent while adding new motion or time progression. This is useful for extending footage that ended too soon.
- First and last frame control: define both the starting and ending visual state, and the model generates the motion between them. This gives precise control over transitions, reveals, and camera moves.
- Reference video workflow: upload a reference clip and the model will follow its motion pattern, timing, and camera behavior while applying your subject and style. This is useful when you have a motion reference but need different content.
- Longer clip planning: the model handles longer generations with better temporal consistency than shorter-duration alternatives, making it a strong choice for clips that need to hold together over more than a few seconds.
These strengths should shape the prompt and the review process. If the strength is reference consistency, upload cleaner references and judge whether the subject stays stable. If the strength is product storytelling, define the product moment before generating. If the strength is speed, use the first outputs to test direction rather than expecting final polish. If the strength is cinematic motion, write camera language instead of generic adjectives.
This is also where many teams waste credits. They choose a model because it is new, not because it fits the job. A better habit is to choose the workflow first. Decide whether the task is exploration, draft, final candidate, prompt research, or campaign review. Then pick the model and settings that match that stage.
Step-by-step workflow
1. Define the asset and channel
Write a one-line production brief before you generate. The line should include the asset type, channel, subject, and purpose. For example: create a 5-second product reveal for a paid social test, create a scene continuation for a longer edit, or create a reference-guided motion clip for a pre-production concept. This prevents the prompt from becoming a vague pile of style words.
2. Choose the source material
If you already have a product photo, portrait, sketch, or reference video, use it only when it improves control. A weak reference can hurt the output more than a strong text prompt helps it. Look for clean subject edges, readable shape, enough background context, and no distracting text or logos. If the source image is not strong enough, create or edit a better reference first.
3. Write the first prompt
The first prompt should be plain and testable. Name the subject, describe the scene, state the action or visual transformation, add camera or composition language, and finish with the intended style. Avoid stacking too many competing instructions. A prompt that asks for macro product photography, handheld documentary realism, anime lighting, floating typography, and a fashion editorial mood at the same time will be difficult to judge.
4. Generate a conservative baseline
The baseline generation is not supposed to be the final winner. It is a diagnostic pass. You are checking whether the model understands the subject, whether the input reference is useful, whether the motion is readable, and whether the output channel makes sense. Save the baseline even if it is imperfect, because it becomes the comparison point for the next variation.
5. Change one variable at a time
When the first output is close, change only one thing. Adjust the camera move, the lighting, the background, the ratio, the duration, or the model. If you change everything at once, you will not know what improved the result. This is the main reason structured workflows beat random prompt experimentation.
6. Review with a checklist
Before keeping an output, check subject consistency, visual clarity, product accuracy, motion readability, composition, background distractions, and publishing fit. For commercial work, also check rights, brand rules, provider terms, and whether the result needs human retouching before release. A beautiful generation that cannot be approved is not a finished asset.
Prompt framework
A reliable prompt for Wan 2.7 has five parts: subject, context, action, style, and constraint. The subject tells the model what matters most. The context gives the scene enough grounding. The action explains what changes. The style defines the visual language. The constraint protects the output from common failures such as unreadable text, product drift, busy backgrounds, or too many actions at once.
Use this structure:
Subject: [main product, character, sketch, scene, or reference]
Context: [environment, lighting, channel, audience, campaign goal]
Action: [movement, transformation, camera behavior, edit instruction]
Style: [commercial, cinematic, editorial, playful, realistic, illustrated]
Constraints: [keep subject consistent, no unreadable text, no logos, simple background]
The framework is intentionally simple. It works because it separates the parts of the brief. If the result fails, you can diagnose the failing part. If the product is wrong, improve the subject line. If the motion is weak, improve the action line. If the mood is off, improve context and style. If the result contains artifacts, tighten the constraints.
Example prompts
Video continuation
Continue from the last frame of the uploaded clip. The car stays parked on the coastal road. Camera slowly pulls back. Sun drops behind the cliff edge. Taillights glow as ambient light fades. Keep the car shape and color consistent. No new subjects entering the frame. Cinematic editorial style.
When you test this inside OpenVideoMaker, change only one variable at a time: the camera speed, the lighting shift, the duration, or the background density. That makes the next result easier to compare with the previous one.
Product reveal
A wireless speaker on a dark wood surface. Camera starts tight on the grille and slowly pulls back to reveal the full product. Soft directional light from the upper right, subtle shadow on the surface. Premium realistic campaign style. Keep the speaker proportions and material detail stable. No text overlays, no logos beyond the product itself.
Reference-guided motion
Follow the camera movement and timing from the uploaded reference video. Replace the subject with a perfume bottle on a marble pedestal. Keep the same slow orbit speed and the same light direction. The bottle should remain centered and stable throughout. Commercial product style. No readable text on the bottle label.
Use cases
Scene continuation
A travel brand has a 10-second drone clip of a mountain lake that ends abruptly. They want to extend it by 5 seconds with the camera continuing its slow pan and the light shifting toward golden hour. They upload the last frame as the first-frame reference, set the duration to 5 seconds, and prompt for a continuation with the same camera direction and a warm light shift. The first pass keeps the landscape consistent but the pan speed is slightly faster than the original. They adjust the action line to specify "same pan speed as the source clip" and the second pass cuts in cleanly.
Scene continuation is one of Wan 2.7's most practical workflows because it solves a real production problem: footage that is too short. The key is matching the camera behavior and lighting of the existing clip so the extension feels like part of the same shot.
Product motion
A consumer electronics company is launching a new headphone. They need a 4-second reveal clip for the product page hero. They upload a clean product render as the first frame, set the last frame to show the headphones at a three-quarter angle, and prompt for a gentle rotation with studio lighting. The model generates the motion between the two frames. The first pass shows the rotation but the ear cushion material looks too smooth. They add "visible texture on ear cushions, matte finish" to the constraints and the second pass is ready for the product page.
Product motion benefits from first-and-last-frame control because you can define exactly where the motion starts and ends. This removes guesswork about the final framing and lets you focus on the quality of the movement between.
Story clip
An animation studio is testing narrative concepts for a short film. They generate a 6-second clip of a character walking through a rain-soaked alley, using a character sketch as the reference and a mood image for the lighting direction. The first generation captures the mood but the character's proportions shift between frames. They add a constraint about keeping the character's height and limb proportions consistent, and the second pass is stable enough to include in the concept reel.
Story clips work best when the prompt focuses on one emotional beat rather than a full narrative arc. A character walking through rain is one beat. A character walking through rain, finding shelter, and watching the sunset is three beats, and the model will handle them better as separate generations.
Social teaser
A beverage brand is launching a new flavor and needs a 3-second teaser for Instagram Stories. They upload a product photo as the first frame, prompt for a slow zoom toward the label with condensation droplets catching the light, and set the duration to 3 seconds. The first generation looks good but the zoom is too fast for the Stories format where viewers tap quickly. They slow the zoom by adjusting the action line and the second pass has the right pacing.
Social teasers need to work in the first half-second. If the key visual does not land immediately, the viewer will tap past it. Front-load the most recognizable element and keep the motion simple.
Reference video variation
A fitness brand has a motion reference clip of a person doing a yoga pose. They want to generate the same motion with a different person in a different setting. They upload the reference video and a new character photo, prompt for the same motion timing and camera angle, and specify a bright studio environment instead of the original outdoor setting. The model follows the reference motion while applying the new subject and environment. The result becomes part of a series where each video uses the same motion pattern with different characters and settings.
Reference video variation is useful when you have a motion pattern that works and want to replicate it across different content. The reference video controls the timing and camera behavior, while the prompt and image references control the subject and environment.
Quality checklist
Use this checklist before you keep a generation:
- Subject accuracy: the main subject should remain recognizable and should not gain unwanted details.
- Composition: the frame should have enough breathing room for the channel where it will appear.
- Motion clarity: if the output is video, the viewer should understand the main movement without replaying the clip.
- Lighting and material: product surfaces, skin, fabric, metal, glass, and shadows should match the intended style.
- Background control: the background should support the subject instead of competing with it.
- Text and logos: avoid relying on generated text unless the model and use case are specifically suited for it.
- Format fit: check ratio, duration, resolution, and crop safety before using the asset in a campaign.
- Legal and brand review: confirm rights, likeness, trademarks, product claims, and provider terms before publication.
The checklist matters because AI media can look impressive while still failing the brief. A clip may have beautiful lighting but show the wrong product detail. An image may look premium but crop badly on mobile. A talking avatar may speak clearly but not match the brand tone. Review each output against the job it was supposed to do.
Common mistakes
The first common mistake is using broad keywords as prompts. Phrases like "best product video" or "cinematic AI ad" describe the category, not the shot. A model needs specifics: what product, what scene, what movement, what style, and what should stay stable.
The second mistake is asking for too many transformations in one generation. If the subject should rotate, the background should change, the camera should zoom, the lighting should shift, and the product should transform, the output may become unstable. Choose the most important change first.
The third mistake is ignoring the source image. Image-to-video and reference-based workflows reward clean inputs. If the source has blur, clutter, strange crop, unreadable labels, or unclear subject boundaries, the output may inherit those problems.
The fourth mistake is treating model choice as a permanent decision. In a multi-model workspace, the point is to compare. Use one model for exploration, another for final polish, and another when a specific input type or style fits better.
The fifth mistake is publishing without review. AI output should be checked for accuracy, rights, brand safety, and channel fit. This is especially important for ecommerce, advertising, education, and any workflow involving likeness or product claims.
Related pages
Use Wan Video Generator when the current article matches your immediate task. Use Image to Video Generator when you need the next step in the workflow. Use AI Image Generator when the brief still needs a strong still frame. Use AI Video Generator when the project needs movement, timing, or camera behavior. Use prompt pages when the hardest part is explaining the desired motion clearly.
FAQ
Is Wan 2.7 the best choice for every project?
No. The best choice depends on input type, output channel, review speed, and creative goal. Wan 2.7 is useful when it fits the workflow described above, but another OpenVideoMaker model or tool may be better when the project needs a different reference type, output style, or iteration pattern.
How should I write the first prompt?
Start with a direct production brief. Name the subject, describe the context, add one main action or transformation, choose the visual style, and include the most important constraint. Keep the first prompt simple enough that you can understand why the output succeeded or failed.
Should I use a reference image?
Use a reference image when it improves control. It is especially helpful for product, character, portrait, and composition-sensitive work. Do not use a weak reference just because the workflow supports one. A clean prompt can outperform a messy reference.
How many variations should I generate?
Generate enough variations to compare direction, but not so many that review becomes random. Three to five focused variations are often more useful than twenty unrelated attempts. Change one variable at a time so the team can understand what caused the improvement.
Can I use outputs commercially?
Commercial use depends on your assets, your rights, the provider terms, and the final content. Review product claims, brand rules, likeness permissions, trademarks, and publishing requirements before using any generated asset in a public campaign.
Final workflow
The best way to use Wan 2.7 is to treat generation as a controlled creative loop. Start with a clear brief. Prepare the input. Write a structured prompt. Generate a baseline. Compare focused variations. Keep the strongest output. Then reuse it as a reference, campaign asset, or next-step input.
For the most direct next step, open Wan Video Generator. If the project needs adjacent workflow support, continue with Image to Video Generator. If you are still choosing between models, start from AI Video Generator or AI Image Generator and compare the model pages that fit your source material.