VoxelPlot

“Klaus” (Academy Award-nominated) used AI to infer consistent 3D surface information from hand-drawn 2D animation. As Sergio Pablos explains, the key problem is surface correspondence. The system must track the same point on the character across different drawings. This creates something similar to a persistent UV map over 2D animation. Once that exists, textures, lighting and other surface information can follow the character’s volume without using a fully animated 3D model. AI could reconstruct fake 3D passes directly from 2D drawings: • normals • depth • position • material masks • UV / surface correspondence Those passes could drive lighting, shadows, materials and FX while preserving the original 2D animation. Pipeline: ✦ Original 2D animation ⬇ AI-generated textures ⬇ Reconstructed 3D passes ⬇ Lighting / shading / FX It is almost the inverse of Arcane: 3D → made to feel 2D Instead: 2D → given the depth and material richness of 3D What concerns me most is that artists who are open to AI-assisted workflows can face harassment or backlash, making some afraid to collaborate even when they are interested. I really hope this changes over time. I'm particularly interested in workflows involving: • color transfer • rendering and relighting • illustrated texture transfer • AI-assisted compositing • hybrid shots combining hand-drawn animation and generated elements • sequences mixing traditional / handcrafted shots, • hybrid shots mixed fully generative shots in final animation This is the link for the full interview with Sergio Pablos: dailymotion.com/video/x7rcg1…

With Wan 3.0 AI anime is getting seriously close to TV quality. You can create sequences like this one that look like an actual seasonal anime. I made a full fantasy fight between a Mage and a Swordswoman — and it genuinely feels anime-ready. Seedance 2.5 keeps characters more on-model. Wan is better at timing, poses and complex layouts. Hailuo H3 sits in between. All three support Vid2Vid and Omni Reference, so you can pass shots and references between them. The possibilities are endless. Interesting times.

GPT-6 Astra + 3D Jutsu I tested this workflow, and I am actually impressed by Astra's understanding of camera work. It seems useful for certain scenes where you need to lock positions and camera movements very accurately. The way to work with this is: 1. Use basic shapes: prisms, spheres and cylinders. 2. No details. 3. Do not add animations; only animate their placement in the scene. 4. Give each subject a different solid color. Rendered in Seedance 2.5 at 480p. When you add detailed elements like animations or geometry that accurately matches the subject, you don't give the model room to imagine, and it just renders directly on top of the scene's geometry.

Since I noticed in my last post that many people love dragon art, I decided to make a dragon animation... (I summoned one)

I finally managed to create an Anime Fight Scene that looks straight out of a real TV show! The latest AI updates and my recent research into animation pipelines are paying off massively, opening up entirely new techniques to explore. Bringing this together required multiple layers and processes, but I’m thrilled with the final result.

If you don’t know what "LATENT SPACE" is, you’re missing one of the most important concepts behind AI Video. Fortunately, Dr. Kit explains it in just 1 minute. Watch this 👇 #DreaminaCPP @dreamina_ai

5. Second Part of the Video Now, since we have a fade to white, we can execute a dissolve effect from white to transition to the second part of the video. In my case, I extracted the transition to white from the first clip and reversed it to recreate the exact same effect and give Seedance the reference. Afterwards, we will use the reference images to create the second part of the video in a single shot. Prompt: REFERENCE MAP: "img" = vid1 (dissolve effect, opening dissolve only) · "27373ff8-c74f-4979-863c-4894866fbae4" = img1 (pale-haired woman in iridescent clothing before a crystalline city) · "c409cd40-4c07-47b3-88dc-650c08456f68" = img2 (ornate silver and gold double-bladed crystal spear, on white) · "783dd18a-c4a2-4311-b5f2-a2b23e679ca8" = img3 (STYLE REFERENCE ONLY, NOT A FRAME: perfectly round water droplets suspended motionless in bright blue air, each throwing sharp starburst flares, dense bokeh field behind) · "b9306d6d-06ab-4b02-b0f9-2c2c3f27ff91" = img4 (STYLE REFERENCE ONLY, NOT A FRAME: faceted ice snowflakes, starburst flares). CRITICAL RULE: img3 and img4 ARE NOT FRAMES OF THIS VIDEO. They are style references only, never reproduced, recreated or shown as a shot. Do not copy their composition, framing or camera angle. Every droplet and snowflake shot is a completely new, originally composed image built only in the spirit of those references. img3 and img4 are never shown full-screen and never appear as an image inside the frame. Cinematic fantasy VFX commercial, Hollywood grade. 15 seconds. Celestial high-key world, strong white backlight, rim light, heavy bloom, prismatic flares, pale blue and silver-white palette turning to brilliant white-gold in the final act. DEPTH OF FIELD IS THE DEFINING LOOK IN EVERY SHOT: extremely shallow focus, wide aperture, only the subject sharp, everything behind it heavy creamy bokeh. 00:00-00:04 - Opens from white and resolves through vid1's dissolve. img1's woman appears in extreme close-up, fully out of focus, a luminous blur. A slow rack focus brings her to perfect sharpness while the camera pushes out and orbits slightly, widening to waist-up. The city behind stays deeply defocused. She appears once. 00:04-00:05:15 - She snaps her fingers and img2's crystal spear materializes floating in the air beside her, four metres long, glowing, arcane glyphs spiralling around it. She reaches out and takes hold of it. 00:05:15-00:07 - Gripping the spear she thrusts it toward the lens, tip aimed straight at camera. Very wide short focal length grossly exaggerating perspective: the tip and her hand loom enormous in the foreground, her face far smaller behind them. The camera dollies fast along the shaft toward her face. Barrel distortion, extreme foreshortening. She casts a spell straight into the lens. 00:07-00:08 - The spell becomes a beam of magical light rushing at the camera: a luminous white-gold shaft built from swirling particles and thin trails of golden smoke, sparks streaming along its length. 00:08-00:10:30 - The beam arrives in a field of perfectly round water droplets in the spirit of img3, hanging completely motionless in the air against luminous blue, each bead crisp and throwing sharp starburst flares, hundreds of them receding into a deep bokeh field. The camera sits inside this suspended cloud and orbits through it at normal speed, weaving between the still drops, parallax sliding them past one another. The magical light threads through the field, its golden particles and smoke curling around each droplet, wrapping them one by one. 00:10:30-00:12 - The light enters the droplets. Each bead ignites from within and turns brilliant gold, no longer water but liquid gold suspended in mid-air, glowing white-hot at the core, then hardens and freezes solid. 00:12-00:13:15 - The frozen golden beads shatter. At the instant of the burst the focus is thrown off completely: the golden snowflake stars in the spirit of img4 are never sharp, appearing only as heavily defocused shapes and glowing bokeh orbs scattering in slow motion across a fully blurred frame. From every burst a thread of white-gold magic escapes and streams outward through the blur. 00:13:15-00:15 - All the threads of golden magic converge and weave together, and the title is revealed out of that light against the already defocused field: bold sans-serif lettering reading exactly "SEEDANCE 2.5", spelled S-E-E-D-A-N-C-E, forming from brilliant white-gold energy with bloom and lens flares. The lettering is the only sharp element in the frame, razor crisp against a completely melted background of golden bokeh, held to the very last frame. Style: fantasy, celestial, photorealistic VFX, god rays, backlit, lens flares, bokeh, slow motion, golden particle magic. Camera fluid, eased. 4K.

4. Generate the Beginning of the Video: 3D Image Panel to Slider Carousel We will use the video generated in step three and attempt, with the image from the first frame, to generate the beginning of the video. Prompt: TITLE CARD, THE FIRST THING THIS VIDEO SHOWS: bold sans-serif lettering fully present and legible in the very first frame at 00:00, floating in the foreground, sharp and in focus from the start. First line reads exactly "SEEDANCE 2.5", spelled S-E-E-D-A-N-C-E. Second smaller line beneath reads exactly "Up to 50 references". Dark charcoal type with the words "50 references" in bright cyan blue. Crisp, sharp-edged, perfectly legible. It sits weightlessly over the moving background and fades out cleanly just before 00:04. High-budget SaaS motion design commercial. Total duration 14 seconds. FIXED TIMING, NON-NEGOTIABLE: reference vid1 starts at exactly 00:04 and plays for ten full seconds, ending at 00:14. Vid1 occupies the majority of this video. It runs from its very first frame to its very last frame as one single continuous playback, in its original order and at its original pace, with all ten seconds present. Its closing seconds play out completely and at full length. The generated portion is only the first four seconds and it must finish by 00:04 so that vid1 begins on time and has room to run whole. THE ENTRY PASSAGE (00:00–00:04), starting from reference img1. It moves briskly and covers its ground within four seconds. Endless bright white space, pale blue, periwinkle and silver-white palette, soft fog, glossy floor with gentle reflections. RACK FOCUS: the passage opens completely out of focus, a soft creamy blur of luminous shapes and glowing colour with no legible detail. Over the first three seconds the focus pulls in gradually and continuously, edges resolving and card borders sharpening until by 00:03 the image is perfectly sharp and holds that way. A smooth cinematic lens rack, eased at both ends. The title lettering stays sharp throughout, unaffected by the blur behind it. It opens on a tight crop of ONE continuous inclined 3D plane built from luminous image cards: only four large cards are visible, and the same grid carries on beyond all four edges of the frame, with further rows above and below and further columns left and right. The plane eases into a slow rotation, tilting toward a more vertical angle while staying inclined, floating weightlessly. A dolly-out then eases in and pulls back steadily from this identical surface. More of the same grid enters the frame in every direction while the four original cards stay visible and shrink, joined by hundreds of smaller cards from the same wall. It remains one single continuous plane at all times, the very same surface throughout, simply seen wider, until the full immense mosaic fills the frame edge to edge, intact and complete, light sweeping across it. A wave of dissolution then sweeps inward from the top edge and the bottom edge simultaneously, cards fading softly into the white void band by band, the wall thinning until a single central horizontal strip remains. That strip enlarges into a BROAD CAROUSEL BELT of large 16:9 panels standing ONE PANEL TALL, a single chain joined only left to right, with open white void above it and open white void and reflective floor below it. Every panel shares the same top edge and the same bottom edge. The belt is slightly tilted in 3D, its ends converging toward vanishing points. The belt travels from left to right, accelerating continuously into a high-speed stream, the camera panning right with steadily increasing angular speed, still gaining momentum as the passage ends. THE HANDOVER INTO VID1: the entry passage ends at 00:04 on a frame matching vid1's first frame exactly in belt speed, camera pan velocity, framing, angle, direction, light and focus, fully sharp on both sides of the join. The movement carries straight through as one continuous gesture and vid1 takes over seamlessly, then runs all the way to its end. The video finishes on vid1's own final frame, the white-out that closes it, arriving only after all ten of its seconds have already played through. The last moment is pure blinding white filling the entire frame, fully blown out edge to edge, held to the very last frame. Panel effects in the entry passage: broad specular reflections rake diagonally across the surfaces like light sliding over polished glass, flaring brighter as they travel and momentarily blowing whole panels out to pure white before easing back. Panel borders glow with a fine luminous rim and prismatic fringing in peach, icy cyan, lavender and pale gold. Bloom is strong yet smooth, holding steady. Colours stay within the scene's own cool spectrum. The entry passage uses strict ease-in and ease-out curves, with the belt phase as one unbroken accelerating ramp. All motion stays fluid and precisely controlled, motion-design grade. Photorealistic 3D render, minimal, elegant, 4K.