Text-to-image vs image-to-video on Kanvora
Use text-to-image to create a new still from a prompt. Use image-to-video after you have an owned starting frame: keep Hailuo for a cheap fixed six-second motion proof or choose Seedance 2.5 when 4-30 seconds, 480p/720p, or native audio is required. These workflows are consecutive tools, not interchangeable models.
Text-to-image
Create a new still from words with GPT Image 2, Nano Banana 2, Seedream 5.0 Lite, Seedream 5.0 Pro, or Seedream 4.5.
- Best fit
- Finding subject, composition, light, treatment, and destination frame.
- Output
- One 1K or 2K still, depending on the model
- Cost
- 1 to 18 credits per attempt, depending on the model and exposed tier
Image-to-video
Animate an owned starting frame with cheap fixed Hailuo or the configurable Seedance 2.5 flagship.
- Best fit
- Motion after composition and subject identity have been reviewed, with model choice driven by duration, resolution, audio, and budget.
- Output
- Hailuo: fixed 6s 768p. Seedance: 4-30s at 480p/720p with native audio
- Cost
- Hailuo: 24 credits. Seedance: 74-1,156 credits
Compare the same decision criteria
This is a handoff comparison, not a contest. Text-to-image establishes the frame; image-to-video performs motion from a frame you already trust.
| Criterion | Text-to-image | Image-to-video | Decision rule |
|---|---|---|---|
| Required input | Text prompt | Starting image plus motion prompt | Without an owned starting frame, use text-to-image first. |
| Output | One 1K or 2K image | Hailuo fixed 6s 768p, or Seedance 4-30s at 480p/720p with native audio | Choose by destination format, not by novelty. |
| Credits | 1 to 18 per attempt | Hailuo 24; Seedance 74-1,156 depending on duration and resolution | Resolve the still before paying for motion, and do not default to the expensive flagship. |
| Primary review | Subject, composition, light, text, and visible facts | Camera move, subject motion, continuity, drift, and the final frame | The second review does not replace the first. |
| Typical stop rule | Stop when the brief depends on a missing reference or exact editable detail | Stop when the start frame is unapproved or the brief needs a longer edited sequence | Move to the workflow that owns the missing requirement. |
| Queue handling | Wait through QUEUED, RUNNING, or UNKNOWN | Allow the longer video timeout and wait through UNKNOWN | Never use a duplicate submission as status recovery. |
On smaller screens, this matrix scrolls inside its own frame. The page itself does not scroll sideways.
A real starting frame and an honest motion handoff
The still below is an official Kanvora Nano Banana 2 generation. The motion brief beside it is a prepared direction for Hailuo, not a rendered clip or a claim about the output it will produce.
AI-generated · Kanvora officialOfficial starting-frame example
A stored 1K Nano Banana 2 still with a clear focal subject and stable horizon.
Nano Banana 2 · 1:1 · Generated Aug 20, 2026Prepared motion direction
A slow, steady camera push toward the telescope while the paper layers move subtly in the sea breeze. Keep the horizon stable, preserve the original composition, and do not add text or new subjects.
One camera move, one environmental motion, explicit preservation, and no new subject.
Open this directionThis comparison does not present its prepared direction as a rendered result. Review the official playable clip on the Hailuo model page, then open the legal creator path and upload a starting frame you are allowed to use.
Use the workflow boundary as the decision rule
A good handoff preserves what is approved and names what may change. These scenarios include action, escalation, and no-action outcomes.
You have a written scene but no starting image.
Signal: Subject placement, camera angle, palette, or destination ratio is still open.
Start with text-to-image.Create the observatory stillHailuo requires a starting frame. Build and review the composition before writing motion around it.
You have an approved still and need one restrained six-second move.
Signal: The source image is owned, the focal subject is clear, and the camera direction fits one clip.
Move to Hailuo.Open the Hailuo motion setupThe live path accepts a private starting-frame upload and a motion prompt, then returns a fixed six-second 768p video.
The destination is a static poster, cover, thumbnail, or product concept.
Signal: Motion adds no required information and the final deliverable is a still.
Keep the still.Open the still workflowA 24-credit video attempt does not improve a static deliverable. Finish typography or layout in the appropriate editor.
You need multiple shots, precise edits, captions, or an editable timeline.
Signal: The output contract extends beyond one generated clip even if Seedance provides native audio and up to 30 seconds.
Escalate to a video editor or production workflow.Review the single-clip setupKanvora does not assemble a multi-shot timeline on this path. Generate only the clip that has a clear role in the larger edit.
Plan the handoff before spending on motion
The still and clip are separate jobs with separate credit holds. Download the approved still, open the Hailuo path, upload that file privately, and keep the motion request narrow enough to review.
Switching and handoff costs
- The starting image must be uploaded as a private Kanvora asset before Hailuo can run.
- A Hailuo attempt costs 24 credits in addition to the still-generation cost.
- Motion may change shape, detail, lighting, or subject identity even when the first frame is approved.
- A six-second clip still needs review, download, and any larger edit outside this single-clip path.
What the comparison cannot prove
- A strong starting frame does not guarantee stable motion or an unchanged subject.
- The prepared motion prompt is not a rendered result or performance benchmark.
- The 768p clip is not an editable timeline, multi-shot sequence, or audio mix.
- An UNKNOWN status means reconciliation is still required, not that a retry is safe.
Finish the frame first. Animate only what has a reason to move.
The still and motion CTAs both carry a legal operation, model ID, and prepared prompt into the real creator. The Hailuo path still requires a private starting-frame upload before submission.