Ben Nash estimates a MiniMax H3 Max film reskin at $1,500-$3,000 or more

• A two-hour 768p reskin pencils out to $576 for one clean output pass, but retries and alternate generations push bennash's estimate to roughly $1,500 to $3,000 or more.
• Each H3 Max clip tops out at 15 seconds, so a clean two-hour pass means at least 480 output jobs, the operational detail behind bennash's reskin estimate.
• A finished AI film can consume far more material than its runtime: MayorKingAI's production tally puts Javi López's 17-minute Nemoris pilot at 12 hours of generated video and more than 7,000 generations.
• Nash sees a future library of work as reskinnable, telling one reply that bennash's reply treats everything already made as material that can be reimagined later.

Fal describes H3 Max as its post-trained MiniMax H3 variant, tuned for prompt adherence and aesthetics. MiniMax's video-generation guide documents text-to-video, first/last-frame, reference-based generation, and video editing, the interfaces a reskin would have to move through one short output at a time.

Nash's headline number is a clean-pass estimate, not a promised production budget.

A two-hour film contains 7,200 seconds. Fal's H3 Max comparison lists 768p at $0.08 per output second, which yields $576 before choosing a second take.

Nash's $1,500 to $3,000-plus band is about 2.6 to 5.2 clean passes. Fal's reference-to-video endpoint also bills reference input beyond its included allowance, and Nash does not specify an input mode, source-asset volume, or shot plan.

H3 Max works in five-to-15-second clips, according to Fal's model comparison, rather than as a single feature-length render.

At the 15-second cap, 7,200 seconds becomes 480 individual outputs. If every output were five seconds, it becomes 1,440.

MiniMax documents three useful control patterns:

• Text-to-video for generating a shot from a description.
• First/last-frame image-to-video for constraining a shot's entrance or exit.
• Reference generation, which can condition on images, video, or audio.

The H3 model card lists output durations up to 15 seconds and native stereo audio, so a reskin budget also leaves open the question of whether sound survives, is regenerated, or is rebuilt separately.

Javi López's Backrooms Nemoris: It Eats Light offers a different production ledger. MayorKingAI's production tally reports 17 minutes of final footage from 12 hours generated, six months of work, more than 7,000 generations, and roughly $20,000 in tokens.

That is about 42 minutes of generated material for each finished minute. The project used Seedance 2.0 and 2.5, Nano Banana Pro, and Seedream 5 through Magnific, not H3 Max, so its spend is a counterexample to a single-pass H3 budget, not a comparison price.

Prompt adherence still creates the surplus. MayorKingAI's comment says longer 15-second generations often need multiple attempts, while MayorKingAI's follow-up describes cutting the strongest pieces from several versions in the editor.

Sound was a separate workload: magnific's production notes says it took 60% of the production time, with more than 13 audio tracks in one scene.

GlennHasABeard's local H3 tests make the short-clip constraint into a shot-design rule. A 15-second run takes a little under an hour on his RTX 3060, he says in GlennHasABeard's hardware reply, after which he upscales in Topaz.

His prompts isolate one visible change per locked-off frame:

• Steam thins and a pale skin spreads across a coffee's surface in GlennHasABeard's coffee test.
• A shadow reaches a chair while its implied lighthouse remains off-screen in GlennHasABeard's shadow test.
• A balloon exits frame, leaving its thin string as the continuity check in GlennHasABeard's balloon test.
• A flower opens one ring at a time, while the prompt holds every other plant still in GlennHasABeard's flower test.

He says he queues a week's prompts locally and gets about a clip an hour in GlennHasABeard's workflow reply. The dollar meter changes on a local box, but the iteration clock remains.

Nash has also published original animated work. bennash's film post introduces The Right to Tremble, a nine-minute sci-fi short inspired by Æon Flux about a city that abolished fear.

He credits the short to Basic Slop in bennash's Basic Slop reply. In a separate reply, he identifies Basic Slop as his own app and says it uses MiniMax H3 in bennash's app note.

He has called AI video a genre that includes animation and hand-drawn work in bennash's genre comment.