Skip to main content

​
Closed-Source Video Models

← Mago video models · Model catalog · Image models →
Closed-source models are third-party models integrated into Mago. They share several properties:
  • Easier to use — fewer settings, faster onboarding.
  • Instruction-based prompts — tell the model what to do, not what the result should look like. (Why?)
  • More expensive — higher credit cost per render than Mago models.
  • Not frame-perfect — frame count, frame rate, character identity, and composition can drift.
  • Stricter moderation — content filters are tighter; VFX content like blood may be blocked.
  • Availability — not available in Relaxed mode (Pro); available in Unlimited mode (Studio). Otherwise they consume credits.
Best fit: short shots, quick results, simple transformations. Less suited to long-form, precision work, or sequences requiring strict consistency.

​
Kling O1 Pro

General-purpose video transformation from Kuaishou.
  • Prompting: instruction-based. E.g. “remove the crowd”, “change time of day to midnight”.
  • Image references: attach up to two reference images, refer to them in the prompt as @image1 and @image2.
  • Elements: attach an Element — a fixed pair of angles (main + secondary) of the same character or object, referenced in the prompt as @Element1 — useful for character continuity.
Example prompts “Replace the couch with the couch from @image1.” “Add the character from @image1 walking through the scene.” “Change the weather to a heavy rainstorm.”

​
Kling O3 Pro

Improved version of Kling O1 Pro. Same prompt and reference structure. Generally stronger results.

​
Kling 2.6 Motion Control Pro

Character-replacement specialist. Strong at lip sync and facial expression preservation.
  • Character reference: required. A full-body reference works even for close-ups, though Modify Frame preparation gives better results.
  • Prompting: leave the prompt empty at first. Add one only if specific issues need fixing.
⚠️ Prompt warning — Adding a prompt to Kling 2.6 Motion Control Pro often makes the result worse than leaving it empty. Try empty first.

​
Kling 3.0 Motion Control

Like Kling 2.6 Motion Control Pro but supports Elements — a two-angle reference pair (main + secondary) for the same character. Use when you have two reference views of the same character and want more consistent identity across the shot.

​
Seedance 2.0

ByteDance model. Powerful general transformation with strong creative range.
  • Prompting: instruction-based.
  • Source video reference: refer to it as @video1.
  • Image references: up to 9, as @image1–@image9.
  • Audio generation: optional toggle in Advanced — generates sound for the output.
  • Output resolution: 720p, 1080p (default), or 4K. 4K renders at 1,875 cr/sec.
Example prompts “Use @image1 as the new character. Replace the person in @video1 with this character.” “Remove all people from @video1.”

​
Seedance 2.0 Image-to-Video

ByteDance model. Generates video from a start image — no source video required. It is not in the Render Video model dropdown: open the Global Timeline, click Add in the timeline, then Generate with AI. The window is labelled as a prototype feature and may be unstable. Generations appear in the window’s Generations list, from which you can use one as the input video for a shot.
  • Start image: required. Defines the first frame of the output video.
  • End image: optional. When provided, the clip animates towards it — useful for controlled transitions.
  • Prompt: required. Describe the motion and action you want to see.
  • Duration: 4–15 seconds (default 5).
  • FPS: 1–60 (default 24).
  • Resolution: default 720p. Each option shows its credits per second.
  • Generate audio: optional toggle, on by default.
  • Prompting: instruction-based.
Two quality tiers, switched with the Standard / Fast selector at the top of the settings (an info icon compares the two):
Example prompts “The character smiles and looks to the left.” “Zoom out slowly to reveal the full scene.” “Rain begins to fall across the landscape.”

​
Seedance 2.5 - vid2vid

ByteDance model. Video transformation with reference images and longer clip support.
  • Prompting: instruction-based. Refer to the source track as @video1.
  • Source video reference: required — the timeline track is the video reference.
  • Image references: up to 9, as @image1–@image9.
  • Audio generation: optional toggle in Advanced — generates sound for the output. Defaults off.
  • Output resolution: 480p, 720p (default), or 1080p. 4K is not available on Seedance 2.5.
  • Aspect ratio: Auto (default), 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16.
  • Duration: 4–30 seconds. Seedance 2.0 caps at 15 seconds.
Pricing is based on the integer-rounded clip length, not the raw fraction.
Example prompts “Use @image1 as the new character. Replace the person in @video1 with this character.” “Remove all people from @video1.”

​
Happy Horse 1.0

Alibaba model. Similar in approach to Kling O1 / O3 Pro.
  • Prompting: instruction-based.
  • Image references: up to 5, as @image1–@image5.
  • Output resolution: 720p (default) or 1080p.
Example prompts “Change the weather to summer with sunny skies.” “Replace the background with the cityscape from @image1.”
← Mago video models · Model catalog · Image models →