Why this order works
A common failure when animating artwork is to start in the video generator with a long prompt that tries to describe both the picture and its movement. The model then has to guess what the design looks like while also deciding how it moves, and the output drifts from the original artwork. This workflow splits the job. The design is fixed first, as a file. In the creator's demonstration, the Assistant uses the uploaded design and a short set of keywords to produce a fuller motion prompt. The exact menu names and interaction may change as Magnific updates its interface. Only then does the video model get involved, with the artwork attached as its visual anchor and the prompt as its motion instructions.
That structure is also why the example works as a loop. The design's objects float, tilt and settle rather than traveling across the frame, so a short clip can return close to its starting arrangement. When you choose your own artwork, pieces with separated objects on a clean background will behave better than dense scenes: the model needs room to move each element without redrawing its neighbors.
The workflow, step by step
This workflow is adapted from a demonstration by Instagram creator @kiblatadhi, posted November 6, 2026. ALFW has rewritten the steps and checked the model specifications and credit estimates against Magnific's official Hailuo 3 documentation. The Assistant interaction and exact interface sequence are creator-demonstrated rather than independently reproduced by ALFW.
*Step 1 — Finish the static design first.*
Export the poster or graphic at the aspect ratio you intend to publish. Every flaw in the still will move. Soft edges, stray artifacts and crowded corners all become more visible once they are in motion, and no prompt fixes a weak source image.
*Step 2 — Upload the design to a Magnific Space.*
In the Reel, the creator prepares the asset and uploads it to the Spaces dashboard first, so the same file can feed both the Assistant and the generator. Working from one uploaded asset, rather than re-uploading at each step, keeps the generator anchored to exactly the artwork you approved.
*Step 3 — Open the Assistant and give it keywords, not a brief.*
The creator connects the asset, then types a short set of style words. The creator's demonstrated input was: "clean, tactile 3D aesthetic, typography-led colour play, fluid transitions, satisfying seamless loop". The Assistant converts those keywords into a full generation prompt. Keep your own keyword list equally short and concrete: name the material feel, the kind of movement and whether you want a loop. Vague words produce vague motion.
*Step 4 — Move to the Video Generator and choose MiniMax H3.*
Select the H3 model, then connect both inputs the earlier steps produced: the Assistant's prompt and the original design asset. Attaching the design is the step that holds the art direction; a prompt alone will get you a similar scene, not your scene.
*Step 5 — Set a short duration and generate.*
The example aims at a seamless loop for promotion, which favors the short end of H3's 5 to 15 second range. Generate, watch where objects deform or drift, adjust the keywords rather than rewriting everything, and regenerate. In the Reel, the finished animation has the poster's elements floating and transitioning smoothly in 3D, close enough to the source design to read as the same piece, now moving.
*Step 6 — Choose the variant like it matters, because it does.*
Before the final render, check which Hailuo 3 option and reference configuration you are using. Magnific's published averages at list price are approximately 290 credits for Hailuo 3 Max Turbo, 595 for Hailuo 3 Max, 1,080 for standard Hailuo 3, and 1,620 for a generation using a reference video. These are averages, not guaranteed charges; actual consumption varies with clip length. Iterate on Max Turbo while the loop is still wrong. Save the standard option, or the reference-video option at about 1,620, for the version you intend to publish.
Where it breaks
Three limits are worth knowing before you spend credits. First, H3 invents the motion. You can steer it with keywords and the source image, but you cannot keyframe an object's path, so anything that must move in an exact arc belongs in an editor. Second, text in the artwork is at risk: lettering that is crisp in the still can soften or warp once animated, which is why the example leans on objects rather than copy. If the poster's words carry the message, keep the clip short and check the text frame by frame. Third, "seamless" is an aspiration, not a setting. A loop that returns near its first frame still needs checking at the join. If the transition looks wrong, try regenerating with more explicit instructions about the ending position and movement before rewriting the entire prompt.
None of that undercuts the basic technique. For the common job, a finished graphic that needs to move on a feed, this is faster than rebuilding the piece in a video tool, and the credit table tells you exactly what the convenience costs at each quality tier.
Workflow inspiration: @kiblatadhi, Instagram Reel, November 6, 2026. Rewritten from the demonstrated steps; no part of the original post is embedded or reproduced.
Sources
- [1] @kiblatadhi, Instagram Reel (November 6, 2026, 11:55 UTC; read via the public Reel page at Gate 1 on November 6, 2026: the demonstrated workflow of Spaces upload, the Assistant keyword-to-prompt step with the example keyword set quoted in Step 3, the Video Generator with MiniMax H3 selected, prompt and design asset connected, duration set, and the generated result shown; the creator's separate cost-comparison remark is not relied on in this article because it could not be verified)Read source
- [2] Magnific, "Hailuo 3 (MiniMax H3): AI video in five options", official model page (read first-hand at Gate 1 on November 6, 2026: five H3 options on Magnific; clip length 5 to 15 seconds in one-second steps; 768p and 2K; up to 9 images, 3 videos and 3 audio references per generation; average credits per video at list price of about 290 for Max Turbo, about 595 for Max, about 1,080 for Hailuo 3 and about 1,620 with a reference video, varying with clip length)Read source