How would Balenciaga look 38k years later
Made via MagoStudio local :D

Ok here we go!

Disco WarpFusion v0.9.1:

Turbo and quality loras support
6-step turbo lora (vs 30 step vanilla - 5x speedup) and aesthetic quality lora (both early stage previews)

Optional Inference Optimizations
- compile CLIP and secondary
- batch CLIP cuts
- lower about +30% speedup with all enabled

and other tweaks + stuff from stable warp notebook, more details here - patreon.com/sxela/posts/di… (public)

the best thing ofc is that you can cut the render times on 40% with all the modern bells and whistles in just one codex prompt, and then leave a parameter sweep overnight to find the best settings

Looks like discodiffusion warp is getting a second coming
Stay tuned, like those loras 😎

We're living up to the promise (unlike others, wink wink) and open-sourcing Mago Studio today

Use you local GPU via @ComfyUI , or a @vast_ai / @runpod instance.

Prompt a chatbot to use your preferred model weights or workflows - no problem, update anything to your taste - it re is already a solid foundation based on the time-tested MagoStudio webapp.

LMK what features or models you'd like to see in the local version next in the comments

The code is here - github.com/MagoStudio/Mag…

Win, Max, Ubuntu builds (if you can't run npm run dev :D) - github.com/MagoStudio/Mag…

doom.sxe.la

someone launched DOOM on subwaysurfer

Updated REGIS skill with findings from doom runner and railway runner

What's new:

Genre and project tooling
- Add regis init --genre survival|runner; existing and unspecified projects retain the survival default.
- Runner briefs cover candidate roles, lane/action rules, ic adaptations, equipment safeguards and end-of-round policy.
- Add runner design, implementation, track-generation, fairness and novice/average/expert playtest references.
- Require mobile swipe and combined-action checks where those controls apply.

Shared production workflows
- Dedicated visual-style and renderer selection, sprite/atlas production, 3D asset benchmarks and equipment-comparison guidance.
- Localization workflow covering game text, layout and release presentation.
- Stronger streaming/loading, runtime performance, native UI and promo-production guidance.
- Tool and skill selection guidance without requiring a sibling Runner checkout.

Illustrated example
- Include five authorized screenshots and a provenance manifest from the authored Railway Runner example. They illustrate a real renderer and declared capture conditions; they are not evidence of human playtesting, physical-device coverage or balance.

Localized social link previews
- Localized preview metadata delivered in the initial HTML rather than relying on client-side JavaScript.
- Native or source-consistent share artwork connected to the actual game's presentation.
- Explicit social-preview checks as part of game delivery and release preparation.

github.com/Sxela/regis

magojeetsu is in -> you can create videos from idea to final cut in natural language, and iterate/edit the project in the chat too!

comment to get early access!

Love the model, it feels like a stablediffusion era for ref2va. Other great models that were good at t2v or i2v were just meh at ref2v, but this one hit the sweet spot!

we're building a local opensource director tool (from idea to final video with character, env and prop assets) built around Minimax h3 and it truly saves you a lot of time managing references and their prompt adaptation

It even kindly shows you the savings vs cloud rendering :D

Stay tuned, code coming in a few weeks!

Brittania Survivors

Behold, the Ultima Online -> Vampire Survivors port is here.

Let the nostalgia in with the original Renaissance client monsters, items, spells, and locations!

Made with REGIS - github.com/Sxela/regis

Opensource ofc, code here - github.com/Sxela/brittani…

Weupport: bs.sxe.la

REGIS: Reverse Engineer Games Into Survivors

Behold, here's the skill used to make the Allods Survivors, and also the Ultima Online and Fallout2 ports.

github.com/Sxela/regis

Just install the skill for your preferred chatbot harness, point it to the downloaded game files, pt, and it will make a vampire survivors-inspired gameplay
@OpenAI codex, @claudeai code, and others are supported

End to end, with playtest, balancing, and a promo video (with ai generated cutscenes if you have local @MiniMax_AI via @ComfyUI )

- Source discovery - build from the game's identity, with evidence for each mapping.

- Complete depth at any content scope - fewer maps or spells still make a complete game.

- Distinct classes and useful builds - give every run meaningful decisions.

- Complete asset lifecycles - reproduce more than icons and walking sprites.

- Native UI and worn equipment - preserve the screens and silhouettes players recognize.

- Practical item handling - compare, equip and clear inferior loot without losing tradeoffs.

- One source-supported companion - bring a useful ally and their personality into the run.

- Desktop and mobile controls - choose inputs for the player and test them in the real game.

- Playtests that catch regressions - measure navigation, performance and earned progression.

- Gameplay-led promos - show real features and verify the exported result.

- Reproducible delivery - prove the player's setup works beyond the development checkout.

- Portable harness installation - carry the same workflow into your chosen agent.

example of its results: allods.sxe.la

Bookmark for new updates, reply which your childhood games you'd like to see reanimated into survivors mode!

github.com/Sxela/regis

Behold, le new version of Allods Survivors!

Imagine Vampire Survivors or Deep Rock Galactic: Survivor in the world of Allods

Made via REGIS: Reverse Engineer Games Into Survival skill

Play today here: allods.sxe.la

Code here: github.com/Sxela/allods-s…

like, share s the opensource REGIS skill to give a new life to your favorite games, and comment which games you want me to do next!

Hosted web-version with mobile support has arrived - allods.sxe.la

(mobile-friendly builds are here - github.com/Sxela/allods-s… )

ui update :D

Felt bored and built Crimsonland in the world of @allods

Loot, a shop, waves of enemies, level-ups, and an absolutely cursed cringe UI included.

The trailer was also made by ChatGPT, so send your complaints to @OpenAI

Electron builds for Windows and Linux are available, ode. You can also run it in your browser using nodejs.

Used d2survivor as a reference and the Reverse Engineer Anything skill - github.com/morluto/rea

Assets are from here - allods2.eu

P.S. Ultima Online is next 😎

github - drop a star ⭐ - github.com/Sxela/allods-s…

Downloads - Windows & Linux - github.com/Sxela/allods-s…

Done in one prompt with @ComfyUI router api and local mago app. 5$ api costs, that's crazy low compared to some other cloud services <wink wink>

Local generation runs on @MiniMax_AI h3 (my favorite local video model so far btw!). On a 5090, a 10-second clip takes 3 minutes. You need 16 GB of VRAM, Windows and 60 GB of disk.

Local generation runs on MiniMax H3. On a 5090, a 10-second clip takes 3 minutes. You need 16 GB of VRAM, Windows and 60 GB of disk.

Ok, time for some good news for the local AI gang.

We've just shipped a desktop version of Mago Studio that runs on your own GPU. Give it an idea or a script and it makes the film. Rendering on your machine is free.

Link below.

Open-source coming soon, as usual, so like and e link 😎

And the best part of it is that it's open-source. Spend a few hours to add local image generation and editing via @ComfyUI et voila!

supports qwen, flux, hidream, sdxl inpaint.

the fork is here - github.com/Sxela/photocra…

It's time to finally reveal the hidden gem of Mago - lo and behold, the Long Form. Now you can go from idea or script to full edit in one go, with full control and simple iteration ui Mago's been famous for.

Say goodbye to yolo-slots-machine script2video workflows you can't in @mago_studio_ai today, and locally (read as: fully offline and open source) soon, hehe.

Having lots of fun with this character

120 video renders later (10hrs on 5090, 67$ saved vs fal ai :D)

Detective Whiskers - the case of the missing sardine

Details in comments

if llms learn to count like that no wonder they can't count strawberries 😁

almost in one prompt with local mago

No ai vertical drama, only vertical chill

Netflix has just dropped ID-V2V two months ago lmao

Is this intentionally misleading? The speedup from the VDN is 3x, the rest comes from going from 50 to 8 steps and from 1 to 8 GPUS.

you need five

had the same idea -

Claude throwing a tantrum after I've sent yet another piece of code to review @ClementDelangue

policy in the comments

microducks are cool but hear me out

GLM-5.3-flash (320b): gets released
@Alibaba_Qwen Qwen-3.8-flash-next (125+51b A6b):

If you used WarpFusion back in the days and have missed on the VibeWarp -

github.com/Sxela/VibeWarp

WarpFusion, unpacked from the notebook

VibeWarp is WarpFusion v0.37 consolidated into an installable Python package with a local web UI — no Colab, no runtime git clone, every or pinned.

It feeds each frame of a video through Stable Diffusion and other image edit models, warping the previous render forward along optical flow so every frame builds on the last instead of being redrawn from scratch. The result is painterly and alive — it drifts, dissolves and reconstitutes.

That texture is the point, not an artefact of it. A video model gives you a clean, coherent
clip; this gives you something that looks made by hand.
This is a cumulative overview of everything in v0.7.x. If you're arriving now, this is the
whole surface.

The render engine
Stable Diffusion 1.5 and SDXL, with the full WarpFusion loop: optical-flow warping, consistency masks, reconstructed-noise mode, and per-frame scheduling of essentially every knob.

- ControlNet — the full SD1.5 and SDXL stacks, multi-net with per-net weights and scheduling. Annotators are built in: depth (Midas / Zoe / Depth-Anything), softedge, scribble, canny, MLSD, normalbae, segmentation, lineart and lineart-anime, shuffle, tile, ip2p, temporalnet, inpaint, and OpenPose / DWPose with per-part body, hand and face detection.

- AnimateDiff — SD1.5 v1/v2/v3 plus SDXL / HotshotXL motion modules, sliding context windows, seam handling and stylized reinjection. The motion module is fully vendored.

- LORA — A1111-format <lora:name:weight> parsing in prompts, plus per-frame weight scheduling.

- IP-Adapter — a canonical model catalog driving a dedicated editor, with the right image encoder (ViT-H or bigG) picked from the adapter itself. Multiple adapters at once, per-frame reference sources, and the full ComfyUI layer-weight preset set including style transfer, composition and the precise variants.

- Gradient guidance — pixel-space (LPIPS + MSE) and latent-space guidance steering sampling toward a chosen temporal target, RMS-clamped, with the target optionally noised to the current sigma.

- Consistency masks — per-component weights, blur, dilate, softening, all schedulable.

- Prompts — per-frame keyframes, multi-prompt blending with weights ( a:0.7 | b:0.3 ), and {caption} interpolation from BLIP frame captioning.

- Content-aware scheduling — scene-change detection (RMSE / LPIPS) driving steps, CFG, style strength and flow blend.

- Plus FreeU, softcap, background masking, tiled VAE, colormatch, deflicker, RealESRGAN upscaling, audio preservation, and fixed_code / reconstructed / pingpong noise modes.
Speed. A multiscale sampler runs early steps at reduced resolution and later steps at full. Optional U-Net caching (DeepCache, First Block Cache) reuses work across denoiser evaluations, and compile_unet torch.compiles the U-Net, ControlNets and VAE paths. A tiled sampler handles resolutions that won't fit otherwise.

Image-edit models

FLUX.2 Klein Edit (4B / 9B)
HiDream-O1 Edit
Qwen Image Edit 2511 (+ GGUF)
Mage-Flow Edit (4B, + turbo)

Alongside the diffusion pipeline, VibeWarp drives four instruction-edit models through an external ComfyUI server over HTTP — nothing is vendored, and live progress is relayed back into the UI.

All four share an ordered Reference Images editor — the first image is always sent, later slots take raw or temporal frames or an uploaded style reference via drag-and-drop.

Temporal contact sheets (experimental) tile several frames into one image and edit them together, so the model sees a frame's neighbours instead of each frame in isolation, then split the result back into frames. Adaptive or fixed layouts, a configurable gutter, and an optional per-sheet instruction. Works across all four edit families, snapping to each model's preferred resolutions.

The interface

A local web UI (Svelte + FastAPI) at http://localhost:7860, plus a full CLI (github.com/Sxela/VibeWarp…). The form is generated from the config itself, so it never drifts from what the engine accepts.

Render. Settings grouped by how you actually use them — what you set once per project, what you change every run, and what you touch once or never. The input video is probed on the spot, so the resolution and frame count shown are what the render will genuinely produce, not an estimate. Queued rendering, live progress and logs, and full config import/export. The ControlNet panel scans your model directory and tells you which checkpoints are actually present before you start.

History & Comparison. Every past run, inspectable frame by frame and layer by layer: the init frame, warped init, processed consistency mask, each ControlNet's source and detected map, the diffusion input, and the output — side by side on a frame slider, so you can see exactly what a ControlNet was fed and what it produced.

• Loop playback — press play on the frame stepper to judge motion without assembling a video first, including part-way through a render.

• Resume — cancelling never discards rendered frames. Continue a run from its first missing frame with the previous stylized frame restored, or assemble a partial video at any time and play it in place.

• Compare — Ctrl/Cmd-click any two runs to diff every saved setting, including settings one run predates entirely.

• Load settings — pull any past run's config back into the form to iterate on it. Supports vanilla WarpFusion settings.

• Errors and settings search - you now see which input field failed validation and can just search for a setting you need.

The run gallery stays usable with hundreds of runs: thumbnails are cached per run rather than served as full renders, and a run still rendering updates its frame count live.

Hope you like it!

For more updates follow me on patreon - patreon.com/c/sxela

political satire: check

surreal @Toyota basketball game: check @bfl_ai

voice prompt + lipsync: check

A young woman in a bright yellow dress stands in the middle of a beautiful forest clearing on a sunny day. She is centered in the frame, facing the camera directly in a medium shot. The background consists of lush green trees, soft grass, and warm ing through the leaves. The camera remains completely static throughout the shot.

The woman maintains steady eye contact with the camera and speaks naturally with clear, accurate lip movements, realistic facial expressions, subtle head motion, and synchronized mouth animation. Her speech is fluent and emotionally neutral.

She says in English:
"Hi, this video was generated by the FLUX 3 neural network."

Ultra-realistic cinematic video, photorealistic skin, natural lighting, high facial detail, perfect lip synchronization, stable framing, 4K, 24 fps, no subtitles, no background music, no camera movement, no cuts.

epicness test: check

SCENE 1 (0-3 seconds) Dynamic opening shot: Camera rapidly pushes through dark forest atmosphere with mystical particles floating in air. Arcane-style painterly textures with dramatic rim lighting. A pack of six grotesque goblins with exaggerated features arling and brandishing crude weapons. Their movements are jerky and aggressive. Heavy percussion soundtrack begins with tension-building strings.
SCENE 2 (3-4 seconds) Quick cut to close-up: Red cloak billowing dramatically in slow-motion as female elf warrior spins into frame. Her face shows fierce determination with glowing eyes. Arcane-style cel-shaded animation with hand-painted texture details. The single one-handed sword catches light with magical energy trails. Sound: Whooshing fabric mixed with metallic sword ring, music intensifies with epic brass.
SCENE 3 (4-6 seconds) Wide action shot: Elf launches forward in fluid combat motion. She slashes through first two goblins with graceful spinning attack - their bodies dissolve into stylized shadow particles. Camera follows her movement with dynamic motion blur and speed lines typical of Arcane animation style. Three remaining goblins circle around her. Sound: Rapid sword slashes (whoosh-clang), goblin shrieks, intense orchestral percussion with electronic undertones. SCENE 4 (6-8 seconds) Fast-paced montage of combat exchanges: Elf parries goblin club with sword spark effects, kicks another goblin backward, red cloak swirls creating striking visual composition. Arcane-style impact frames and exaggerated motion. Goblins attack from multiple angles but she deflects each strike with supernatural agility. Sound: Multiple impact sounds (metal clashes, body hits), grunts and battle cries, driving electronic-orchestral hybrid music reaches climax.
@bfl_ai