BREAKING NEWS
LATEST POSTS
-
Marigold – repurposing diffusion-based image generators for dense predictions
Marigold repurposes Stable Diffusion for dense prediction tasks such as monocular depth estimation and surface normal prediction, delivering a level of detail often missing even in top discriminative models.
Key aspects that make it great:
– Reuses the original VAE and only lightly fine-tunes the denoising UNet
– Trained on just tens of thousands of synthetic image–modality pairs
– Runs on a single consumer GPU (e.g., RTX 4090)
– Zero-shot generalization to real-world, in-the-wild imageshttps://mlhonk.substack.com/p/31-marigold
https://arxiv.org/pdf/2505.09358
https://marigoldmonodepth.github.io/
-
Runway Aleph
https://runwayml.com/research/introducing-runway-aleph
Generate New Camera Angles
Generate the Next Shot
Use Any Style to Transfer to a Video
Change Environments, Locations, Seasons and Time of Day
Add Things to a Scene
Remove Things from a Scene
Change Objects in a Scene
Apply the Motion of a Video to an Image
Alter a Character’s Appearance
Recolor Elements of a Scene
Relight Shots
Green Screen Any Object, Person or Situation
-
Mike Wong – AtoMeow – A Blue noise image stippling in Processing
https://github.com/mwkm/atoMeow
https://www.shadertoy.com/view/7s3XzX
This demo is created for coders who are familiar with this awesome creative coding platform. You may quickly modify the code to work for video or to stipple your own Procssing drawings by turning them into
PImageand run the simulation. This demo code also serves as a reference implementation of my article Blue noise sampling using an N-body simulation-based method. If you are interested in 2.5D, you may mod the code to achieve what I discussed in this artist friendly article.Convert your video to a dotted noise.
-
Aitor Echeveste – Free CG and Comp Projection Shot, Download the Assets & Follow the Workflow
What’s Included:
- Cleaned and extended base plates
- Full Maya and Nuke 3D projection layouts
- Bullet and environment CG renders with AOVs (RGB, normals, position, ID, etc.)
- Explosion FX in slow motion
- 3D scene geometry for projection
- Camera + lensing setup
- Light groups and passes for look development
-
Tauseef Fayyaz About readable code – Clean Code Practices
𝗛𝗲𝗿𝗲’𝘀 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗺𝗮𝘀𝘁𝗲𝗿 𝗶𝗻 𝗖𝗹𝗲𝗮𝗻 𝗖𝗼𝗱𝗲 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗲𝘀:
🔹 Code Readability & Simplicity – Use meaningful names, write short functions, follow SRP, flatten logic, and remove dead code.
→ Clarity is a feature.
🔹 Function & Class Design – Limit parameters, favor pure functions, small classes, and composition over inheritance.
→ Structure drives scalability.
🔹 Testing & Maintainability – Write readable unit tests, avoid over-mocking, test edge cases, and refactor with confidence.
→ Test what matters.
🔹 Code Structure & Architecture – Organize by features, minimize global state, avoid god objects, and abstract smartly.
→ Architecture isn’t just backend.
🔹 Refactoring & Iteration – Apply the Boy Scout Rule, DRY, KISS, and YAGNI principles regularly.
→ Refactor like it’s part of development.
🔹 Robustness & Safety – Validate early, handle errors gracefully, avoid magic numbers, and favor immutability.
→ Safe code is future-proof.
🔹 Documentation & Comments – Let your code explain itself. Comment why, not what, and document at the source.
→ Good docs reduce team friction.
🔹 Tooling & Automation – Use linters, formatters, static analysis, and CI reviews to automate code quality.
→ Let tools guard your gates.
🔹 Final Review Practices – Review, refactor nearby code, and avoid cleverness in the name of brevity.
→ Readable code is better than smart code. -
Mark Theriault “Steamboat Willie” – AI Re-Imagining of a 1928 Classic in 4k
I ran Steamboat Willie (now public domain) through Flux Kontext to reimagine it as a 3D-style animated piece. Instead of going the polished route with something like W.A.N. 2.1 for full image-to-video generation, I leaned into the raw, handmade vibe that comes from converting each frame individually. It gave it a kind of stop-motion texture, imperfect, a bit wobbly, but full of character.
-
Microsoft DAViD – Data-efficient and Accurate Vision Models from Synthetic Data
Our human-centric dense prediction model delivers high-quality, detailed (depth) results while achieving remarkable efficiency, running orders of magnitude faster than competing methods, with inference speeds as low as 21 milliseconds per frame (the large multi-task model on an NVIDIA A100). It reliably captures a wide range of human characteristics under diverse lighting conditions, preserving fine-grained details such as hair strands and subtle facial features. This demonstrates the model’s robustness and accuracy in complex, real-world scenarios.
https://microsoft.github.io/DAViD
The state of the art in human-centric computer vision achieves high accuracy and robustness across a diverse range of tasks. The most effective models in this domain have billions of parameters, thus requiring extremely large datasets, expensive training regimes, and compute-intensive inference. In this paper, we demonstrate that it is possible to train models on much smaller but high-fidelity synthetic datasets, with no loss in accuracy and higher efficiency. Using synthetic training data provides us with excellent levels of detail and perfect labels, while providing strong guarantees for data provenance, usage rights, and user consent. Procedural data synthesis also provides us with explicit control on data diversity, that we can use to address unfairness in the models we train. Extensive quantitative assessment on real input images demonstrates accuracy of our models on three dense prediction tasks: depth estimation, surface normal estimation, and soft foreground segmentation. Our models require only a fraction of the cost of training and inference when compared with foundational models of similar accuracy.
-
-
Embedding frame ranges into Quicktime movies with FFmpeg
QuickTime (.mov) files are fundamentally time-based, not frame-based, and so don’t have a built-in, uniform “first frame/last frame” field you can set as numeric frame IDs. Instead, tools like Shotgun Create rely on the timecode track and the movie’s duration to infer frame numbers. If you want Shotgun to pick up a non-default frame range (e.g. start at 1001, end at 1064), you must bake in an SMPTE timecode that corresponds to your desired start frame, and ensure the movie’s duration matches your clip length.
How Shotgun Reads Frame Ranges
- Default start frame is 1. If no timecode metadata is present, Shotgun assumes the movie begins at frame 1.
- Timecode ⇒ frame number. Shotgun Create “honors the timecodes of media sources,” mapping the embedded TC to frame IDs. For example, a 24 fps QuickTime tagged with a start timecode of 00:00:41:17 will be interpreted as beginning on frame 1001 (1001 ÷ 24 fps ≈ 41.71 s).
Embedding a Start Timecode
QuickTime uses a
tmcd(timecode) track. You can bake in an SMPTE track via FFmpeg’s-timecodeflag or via Compressor/encoder settings:- Compute your start TC.
- Desired start frame = 1001
- Frame 1001 at 24 fps ⇒ 1001 ÷ 24 ≈ 41.708 s ⇒ TC 00:00:41:17
- FFmpeg example:
ffmpeg -i input.mov \ -c copy \ -timecode 00:00:41:17 \ output.movThis adds a timecode track beginning at 00:00:41:17, which Shotgun maps to frame 1001.
Ensuring the Correct End Frame
Shotgun infers the last frame from the movie’s duration. To end on frame 1064:
- Frame count = 1064 – 1001 + 1 = 64 frames
- Duration = 64 ÷ 24 fps ≈ 2.667 s
FFmpeg trim example:
ffmpeg -i input.mov \ -c copy \ -timecode 00:00:41:17 \ -t 00:00:02.667 \ output_trimmed.movThis results in a 64-frame clip (1001→1064) at 24 fps.
FEATURED POSTS
-
The Hollywood VFX Jobs Most At-Risk From AI
https://www.hollywoodreporter.com/business/business-news/ai-hollywood-workers-job-cuts-1235811009/
Over the next three years, it estimates that nearly 204,000 positions will be adversely affected.
In November, former Dreamworks founder Jeffrey Katzenberg said the tech will replace 90 percent of jobs on animated films.
Roughly a third of respondents surveyed predicted that AI will displace sound editors, 3D modelers, re-recording mixers and audio and video technicians within three years, while a quarter said that sound designers, compositors and graphic designers are likely to be affected.
Roughly a third of respondents surveyed predicted that AI will displace sound editors, 3D modelers, re-recording mixers and audio and video technicians within three years, while a quarter said that sound designers, compositors and graphic designers are likely to be affected.
AI tools may increasingly be used to help create images that can streamline character design and storyboarding process, lowering demand for concept artists, illustrators and animators.
According to the study, the job tasks most likely to be impacted by AI in the film and TV industry are 3-D modeling, character and environment design, voice generation and cloning and compositing, followed by sound design, tools programming, script writing, animation and rigging, concept art/visual development and light/texture generation.
-
Google Street View Hyperlapse
All Google Street View imagery captured using hyperlapse.tllabs.io. Source code available at github.com/TeehanLax/Hyperlapse.js.
Read the full story: teehanlax.com/labs/hyperlapse/
-
What Is The Resolution and view coverage Of The human Eye. And what distance is TV at best?
https://www.discovery.com/science/mexapixels-in-human-eye
About 576 megapixels for the entire field of view.
Consider a view in front of you that is 90 degrees by 90 degrees, like looking through an open window at a scene. The number of pixels would be:
90 degrees * 60 arc-minutes/degree * 1/0.3 * 90 * 60 * 1/0.3 = 324,000,000 pixels (324 megapixels).At any one moment, you actually do not perceive that many pixels, but your eye moves around the scene to see all the detail you want. But the human eye really sees a larger field of view, close to 180 degrees. Let’s be conservative and use 120 degrees for the field of view. Then we would see:
120 * 120 * 60 * 60 / (0.3 * 0.3) = 576 megapixels.
Or.
7 megapixels for the 2 degree focus arc… + 1 megapixel for the rest.
https://clarkvision.com/articles/eye-resolution.html
Details in the post
-
Black Forest Labs released FLUX.1 Kontext
https://replicate.com/blog/flux-kontext
https://replicate.com/black-forest-labs/flux-kontext-pro
There are three models, two are available now, and a third open-weight version is coming soon:
- FLUX.1 Kontext [pro]: State-of-the-art performance for image editing. High-quality outputs, great prompt following, and consistent results.
- FLUX.1 Kontext [max]: A premium model that brings maximum performance, improved prompt adherence, and high-quality typography generation without compromise on speed.
- Coming soon: FLUX.1 Kontext [dev]: An open-weight, guidance-distilled version of Kontext.
We’re so excited with what Kontext can do, we’ve created a collection of models on Replicate to give you ideas:
- Multi-image kontext: Combine two images into one.
- Portrait series: Generate a series of portraits from a single image
- Change haircut: Change a person’s hair style and color
- Iconic locations: Put yourself in front of famous landmarks
- Professional headshot: Generate a professional headshot from any image
