ClipCanva

MiniMax H3 Max Workflow: Fast 480p Drafts, First and Last Frames, Then 2K Keepers

Use MiniMax H3 Max for 5–15s 480p/768p text and first-last-frame drafts, then promote keepers to MiniMax H3 2K. Specs, credits, and a creator checklist.

September 4, 2026ClipCanva Editorial

MiniMax H3 Max Workflow: Fast 480p Drafts, First and Last Frames, Then 2K Keepers

MiniMax H3 Max is the speed tier of MiniMax H3, not a replacement for the 2K model. MiniMax’s video generation guide describes H3 Max as jointly released with fal.ai, post-trained on MiniMax H3, and currently limited to text-to-video and image-to-video. Output is 480P or 768P, 5–15 seconds (integers only). Reference generation is listed as coming soon. Use Max to burn cheap first-and-last-frame tests. Promote only the stills and prompts that survive to MiniMax H3 at 2K. ClipCanva is not affiliated with MiniMax, fal, Runway, Canva, Kling, or Google, and ClipCanva does not currently host the H3 Max identifier. Write the packet here, draft on Max, then generate keepers on ClipCanva’s MiniMax H3 page.

Official facts for H3 Max and nearby desks

Source Public claim Workflow implication
MiniMax video generation guide Two models: MiniMax H3 (MiniMax-H3) and MiniMax H3 Max (MiniMax-H3-Max). H3 Max is jointly released with fal.ai, post-trained for speed, and currently supports T2V and I2V only. Reference generation is coming soon. Do not send a 12-file reference pack to Max. That is an H3 job.
Same MiniMax guide H3 output: 768P / 2K, duration 4–15s. H3 Max output: 480P / 768P, duration 5–15s. First/last-frame entry: 0, 1, or 2 images. No image → text-to-video. A 4-second H3 clip is legal. A 4-second Max clip is not. Count frames before you pick the model.
MiniMax pay-as-you-go pricing H3: $0.08/s at 768P, $0.13/s at 2K. H3 Max: $0.05/s at 480P, $0.08/s at 768P. Max bills output only; input images are not billed for now. H3 extra images and input video are billed. Draft at 480P Max. Pay 2K H3 only after the stills and the verb survive.
Runway Dev API changelog (3 Sep 2026) MiniMax H3 Max (h3_max) is on Runway Dev. Text prompt or start-frame image, optional last-frame keyframe. 5–15s at 480p or 768p. promptExpansionMode: disabled, balanced (default), or quality. 5 credits/s at 480p, 8 credits/s at 768p. Runway credits are $0.01 each on the developer portal, so those rates match MiniMax’s $0.05 / $0.08 list. Expansion rewrites the prompt unless you set disabled.
fal: Introducing H3 Max (27 Aug 2026) fal Research post-trained H3 Max on MiniMax H3. fal claims a 5-second clip in under 3 seconds and roughly 35× the throughput of MiniMax’s official H3 endpoint. Preference rankings are fal’s own evaluations. Treat speed numbers as vendor claims. Time your own 5-second 480p test before you promise a client “under 3 seconds.”
Canva AI Video Generator Create a Video Clip, powered by Google Veo-3: one 16:9 clip per prompt, up to eight seconds, with synchronized audio. An 8-second layout clip is not a 12-second first-to-last-frame Max draft.

Access, region, credits, and which identifier your account actually exposes can change. Confirm the live MiniMax, fal, Runway Dev, or ClipCanva panel before you quote 2K, 15 seconds, or reference mode.

Pick the desk before you spend 2K

H3 Max fails when you treat it as “H3, but cheaper, so dump every file in.” MiniMax is explicit: Max is T2V and I2V today. H3 is the model that also takes reference images, videos, and audio.

Job Where it actually runs What you must supply What the model may invent Fail condition
Twelve cheap motion tests H3 Max T2V or I2V at 480P One scene card, one verb, 5–8s World, motion, and (if the live panel writes audio) the bed Sending unlabeled zips; asking Max for 2K
Same bottle, start here, end there H3 Max I2V with first and last frames Two approved stills that share identity The interpolation only A third SKU or a new room between the frames
2K keeper with native stereo ClipCanva MiniMax H3 The surviving stills plus the locked prompt Motion and sound on a 5s or 10s 2K pass Re-prompting from memory instead of the Max winner
Cast + motion plate + voice MiniMax H3 reference mode (API / Hailuo), not Max Named stills, optional clips, optional audio. Cap: 9 / 3 / 3, 12 files In-between motion Pointing that pack at MiniMax-H3-Max or h3_max
8-second 16:9 layout in a design file Canva Create a Video Clip (Veo-3) One prompt, one clip Synced audio inside Canva Asking Canva to hold a 15-second first-to-last path

If the brief is “does this camera move even work,” that is Max at 480P. If the brief is “this label must not melt on the hero,” that is H3 at 2K from approved frames. If you already have a motion plate, that is H3 reference, and Max will not take it yet.

H3 vs H3 Max vs an 8-second layout clip

Dimension MiniMax H3 MiniMax H3 Max Canva Create a Video Clip
Official model name MiniMax-H3 MiniMax-H3-Max / Runway h3_max Veo-3 inside Canva
Modes today T2V, I2V (first/last), reference generation T2V and I2V only One text-to-clip generation per prompt
Resolution 768P / 2K 480P / 768P Not documented as 2K; 16:9 clip
Duration 4–15s 5–15s Up to 8s
List price (MiniMax / Canva) $0.08/s 768P, $0.13/s 2K $0.05/s 480P, $0.08/s 768P Paid Canva plan + credit pool
ClipCanva today Live generator: T2V or 1–2 frames, 5s or 10s, fixed 2K, native audio Not hosted Not hosted
Use it for Keepers, stereo, reference packs Volume drafts and first-last tests A layout clip you will edit in Canva

Do not mix the three in one prompt. A Max draft that looks “good enough at 480p” is a candidate, not a master. Upscaling a rejected still wastes the 2K pass.

Draft-then-keeper workflow

Write the packet once. Change only the variable. Promote only winners.

  1. Name the output. Example: twelve 6-second 9:16 product push-ins, same bottle, four cameras × three lighting tests.
  2. Lock identity as stills. Faces, labels, and lighting are cheaper to fix as images. Use Prompt Ideas for the visual recipe and the AI script generator for the spoken line or the on-screen promise.
  3. Approve two frames when the shot has a start and an end. Same bottle, same table, same type. If the last frame introduces a new cap color, Max will interpolate a product you do not sell.
  4. Run Max at 480P first. Five or six seconds. On Runway Dev, set promptExpansionMode to disabled until the literal prompt is stable. Default balanced rewrites.
  5. Score the test set. Identity hold, camera speed, end-frame match, readable type. Kill the rest.
  6. Promote one packet to H3 2K. Same stills, same verb, 5s or 10s on MiniMax H3. Do not “improve” the prompt while you change resolution.
  7. Human-gate logos, faces, and claims. Then, if you need a second model look, send the same packet through the AI video generator or Compare.

The previous H3 operator note still applies for the keeper pass: MiniMax H3 first-and-last-frame workflow. Max does not replace that page. It sits in front of it.

Operator card for a 12-clip SKU draft

Job: 12x 6s 9:16 product push-ins, same bottle
Draft desk: MiniMax H3 Max / Runway h3_max — 480P, 6s, T2V or I2V
Keeper desk: ClipCanva MiniMax H3 — 2K, 5s or 10s, same frames
Locked: SKU silhouette, label, cap color, table
Variable: camera (push / orbit / tilt-up / static) and lighting (window / studio / dusk)
Frames: approved first still; last still only if the end pose is already shot
Prompt pattern:
  Start on approved still. [camera]. Keep label readable. No extra products.
  End on a clean three-quarter hold. No invented badges or prices.
  Audio: quoted line OR "SFX only, no speech"
Runway-only: promptExpansionMode=disabled on the test set
Test set: 3 tasks (one lighting, push-in only) at 480P Max
Scale: remaining 9 Max drafts only after identity holds
Promote: 2–4 winners to H3 2K, same prompt, do not rewrite
Forbidden: Max reference zips, 2K on the test set, Canva 8s clips as the master

Example, character performance set: lock framing and wardrobe as stills; run Max I2V for three acting verbs (turn, nod, hold); pick the verb that does not morph jewelry; then run that pair of frames on H3 2K. Do not start with 2K.

What Max is allowed to do

Max may: generate from text, animate one start frame, interpolate start-to-end frames, and (on Runway Dev) honor or rewrite the prompt via promptExpansionMode.

Max may not, on MiniMax’s current guide: accept a reference pack of extra stills, motion plates, or audio takes. That is H3. If your Hailuo, fal, or Runway panel later exposes reference-to-video on Max, treat it as a new mode and re-read the live docs. Until MiniMax removes the “coming soon” line, plan as if Max is T2V/I2V only.

Audio: MiniMax documents native stereo for H3. The H3 Max output-spec table lists resolution and duration, not a separate audio toggle. fal’s comparison page says both models write a soundtrack during generation. Runway’s 3 September h3_max note does not mention audio. Confirm the live panel before you promise a silent render or a spoken line.

Creator checklist: 25-minute Max pass

  1. One job. Drafts or keepers. Not both in the same batch.
  2. Write the packet. Viewer promise, locked vs variable, duration, aspect ratio, fail examples.
  3. Choose the identifier. MiniMax-H3-Max / h3_max for drafts. MiniMax-H3 / ClipCanva H3 for keepers.
  4. Respect duration. Max: 5–15s. H3: 4–15s. ClipCanva H3 UI today: 5s or 10s.
  5. Respect resolution. Max: 480P/768P. H3: 768P/2K. ClipCanva H3: fixed 2K.
  6. Attach 0, 1, or 2 frames. Zero is text-to-video. Two is a path, not a collage.
  7. Disable prompt expansion on tests if you are on Runway Dev.
  8. Time one 5-second 480P clip. fal’s “under 3 seconds” is a vendor claim, not your SLA.
  9. Promote winners only. Same stills, same prompt, new model.
  10. Gate type, faces, and claims before anything ships.

For image-led shots, start from image to video after the still is approved. Do not animate the first pretty portrait.

FAQ

What is MiniMax H3 Max? MiniMax’s speed-oriented H3 variant, jointly released with fal.ai and post-trained on MiniMax H3. Official modes today are text-to-video and image-to-video at 480P or 768P, 5–15 seconds. It is not the 2K / reference model.

Is H3 Max on ClipCanva? No. ClipCanva currently connects MiniMax H3 for text-to-video and image-to-video (one opening still, or opening and closing frames) at 5 or 10 seconds, fixed 2K, with native audio. Use Max on MiniMax’s API, fal, or Runway Dev (h3_max). Bring the winning packet back to MiniMax H3.

Should I skip Max and generate everything at 2K? Only if you already know the camera and the stills. At MiniMax list rates, a 6-second 480P Max draft is $0.30; a 6-second 2K H3 clip is $0.78. Twelve untested 2K takes are a credit problem, not a quality strategy.

Can H3 Max take reference video and audio? Not on MiniMax’s current guide. Reference generation is listed as coming soon for Max. Use H3’s reference entry (≤9 images, ≤3 videos, ≤3 audio clips, 12 files total) when you need that pack.

Is Runway h3_max the same as Hailuo MiniMax H3? No. Runway’s help article for Minimax H3 (also called Hailuo 3.0) documents 2K/768p and reference inputs on a Standard plan or higher. The 3 September Runway Dev note for h3_max is the Max identifier: 480p/768p, start frame, optional last frame, no 2K in that changelog. Pick the identifier on purpose.

Sources