Use reference-to-video when a Wan 3.0 brief needs images, motion clips or audio as context. This guide uses APIXO’s route, which also accepts one file or one web link, never both. The LoRA routes do not expose those file/link fields. If you only have a starting image and an optional ending image, check image-to-video first.
wan-3-0-video
The product-video brief below is a planning exercise. It has not been generated for this guide and is not a tested recipe for reproducing every reference detail.
More inputs need clearer roles: Editorial workflow diagram, not a tested model output
Choose the mode that matches your materials
Your starting material: A text brief with no media
Mode to check: text-to-video
Main boundary: A non-empty prompt is required
Your starting material: A first image, optionally a last image
Mode to check: image-to-video
Main boundary: 1–2 images; no video or audio references
Your starting material: Images, video or audio as context
Mode to check: reference-to-video
Main boundary: At least one reference input; media-specific limits apply
Your starting material: A project file or web link as context
Mode to check: Wan 3.0 Video, reference-to-video
Main boundary: One file OR one link; not exposed by the two LoRA routes
For reference-to-video, the published APIXO limits are up to 10 images, 5 videos and 5 audio files. Each reference video or audio file must be 1–15 seconds. Video references total at most 15 seconds, and audio references separately total at most 15 seconds. For a positive fixed duration, output seconds plus all reference-video seconds must not exceed 30.
Check the Wan 3.0 Video documentation for the input requirements. The limits above describe the published API contract checked on October 22, 2026. Confirm which controls your interface exposes before preparing the request.
Start with a deliverable, then assign each asset a role
Suppose you need a transparent bottle on a light table, a slow sideways camera move and a final composition with room for subtitles added in editing. Write those requirements before selecting references.
Asset: Product front view
Intended role: Bottle shape, color and packaging
What to inspect in the output: Geometry, label placement and identity
Asset: Scene image
Intended role: Table, background and lighting
What to inspect in the output: Whether the setting fits the brief
Asset: A five-second motion clip
Intended role: Slow sideways camera movement
What to inspect in the output: Direction, speed and unwanted actions
Asset: One short audio clip
Intended role: Restrained musical atmosphere
What to inspect in the output: Whether sound fits the scene
Asset: One focused project file
Intended role: Audience and visual priorities
What to inspect in the output: Whether the result follows the brief
These roles communicate intent. They do not guarantee that the model transfers only the selected attribute from each asset.
Use images for visible requirements
Choose clear sources with an obvious subject. Additional angles can describe appearance, but conflicting packaging, colors or products make the brief harder to interpret. Remove images that introduce a different version of the subject.
Inspect lettering, logos and geometry throughout the result. A reference image does not guarantee exact reproduction. For source preparation, see the image-to-video starter guide.
Keep video references focused on one action
Choose the segment that shows the required movement. State which action or camera behavior matters and which people, objects or setting are irrelevant.
Trim unrelated cuts, pauses and additional actions, retaining the original file for comparison. Trimming changes both the information supplied and the billable input seconds.
Measure the sum of all reference videos. With ten seconds of video input, a fixed request can ask for at most twenty seconds of output. With fifteen seconds of video input, fixed output can be at most fifteen seconds. Do not apply the headline 30-second output range without checking the reference inputs.
Separate audio context from output sound
Use audio to communicate a desired tone, rhythm or sound direction. A voice reference alone does not establish the speaker’s visual appearance.
The audio reference and the output sound setting do different jobs: one provides context; the other requests whether the output includes sound. Check each when audio matters. Reference audio has no separate input charge, but the output and any reference-video seconds still have costs.
Use one file or one link for context
On Wan 3.0 Video, file_urls and link_urls are alternatives: submit at most one item in either field, never both. Keep the material focused on the audience, product information and visual priorities that are not clear from the media assets.
The service must be able to read the source. A private cloud-drive login page is not a direct asset URL. A document also does not guarantee accurate text, subtitles or product claims in the generated video.
This step applies to standard Wan 3.0 Video. Compare the three APIXO Wan 3.0 routes before carrying it into a LoRA workflow.
Build the reference set one change at a time
To compare reference sets, stay in reference-to-video mode from the first attempt. Begin with one product image and a short written motion brief; an image alone satisfies the reference requirement. Keep the prompt, resolution, fixed output duration and sound setting unchanged when adding the motion clip. Choose an output duration that still fits the combined limit after the clip is added.
Review the product shape and framing, then compare the motion. Add audio context or a file/link in separate attempts only when a requirement is missing. Switching from image-to-video to reference-to-video changes the mode as well as the inputs, so it is not a controlled comparison of the added reference. Generated outputs can also vary between attempts; one better result is not proof that the new asset caused the improvement.
An untested planning prompt for the full asset set:
Preserve the bottle shape and main color from the product image.
Use the scene image for the table and lighting.
The video guides a smooth sideways camera move while the product remains still.
The audio guides a restrained atmosphere.
Add no people or text. Leave space for subtitles to be added in editing.
Reference numbering and syntax must follow the interface you use. This prompt describes the roles of the assets; it is not a tested copy-and-run API request.
Diagnose the first problem before adding more inputs
Symptom: A reference cannot be read
First check: The URL returns the intended media or readable page
Next attempt: Replace expired, sign-in or preview-only URLs
Symptom: The product or clothing changes
First check: Images show conflicting variants or unrelated subjects
Next attempt: Remove conflicts and state the visual priority
Symptom: The action or camera move is wrong
First check: The motion clip contains multiple actions or cuts
Next attempt: Trim to the required movement and explain its role
Symptom: Sound is unsuitable or missing
First check: Audio input and the output sound setting
Next attempt: Check them separately
Symptom: A fixed-duration request exceeds limits
First check: Sum of all video inputs plus output seconds
Next attempt: Keep the sum within 30; video inputs themselves total at most 15
Record the reason for each revision. Changing the images, motion clip, prompt and audio together makes it harder to identify what helped.
Check the cost and the usable result
Reference-video seconds are billed with output seconds. At the standard Wan 3.0 Video 1080p rate of $0.20 per second in the published schema checked on October 22, a hypothetical ten-second output with five seconds of video reference is (10 + 5) × $0.20 = $3.00. This is a calculation, not an observed charge.
See the Wan 3.0 cost guide for the formula and smart-duration pre-charge. Recheck APIXO API Pricing before submitting; these API figures are separate from Studio monthly credits.
Review the result against the brief: subject appearance, movement, sound and space for editing. Keep the settings, materials, accepted seconds and reasons for retries. Open Wan 3.0 Video with a focused reference set, and expand it only when a specific missing requirement justifies the change.
Build with APIXO
Explore image, video, and audio models in one place.
