The Model Library
Every model family follows the same steps: download weights, build or install an engine, use it. This page covers the families, the one-click presets, how Regenerate-Q picks its tile size, and building engines for new resolutions.
Weights → engines
Weights are the neural network itself, downloaded once per family. An engine is that network compiled for your hardware and a specific tile size (CoreML on Apple Silicon). The Model Library's left rail shows each family's state: a green dot means ready; Download weights or Build engine mean there is a step left.
The model families
| Refocus | Fast, consistent CNN upscaler. Used for the standard 4× pass. (Enhance) |
| Regenerate-Q | Streaming, quantized generative upscaler. Adds detail with moderate memory use; the default diffusion model. (Enhance) |
| Frame interpolation | Generates in-between frames to raise the frame rate. (Enhance) |
| Denoise | Temporal, flow-aligned video denoiser. Used by the Enhance pipeline and by Reprocess in Refine. (Enhance and Refine) |
| Segmentation | SAM2 object segmentation. Powers masks in the Refine workspace. (Refine) |
| Depth estimation | Monocular video depth maps, metric or relative. Powers depth masks and the depth overlay. (Refine) |
| Regenerate | The full generative upscaler. Maximum detail, heaviest compute. Listed under Beta and only where it can be built. |
The rail groups families by the workflow they serve: Enhance, Refine, and Beta. The dashboard's Model Library cell shows one readiness meter per group.
Starter presets
Presets follow the workflows. All installs one engine per model family (depth gets one metric and one relative model). Enhance covers upscaling and cleanup: Refocus, Regenerate-Q, frame interpolation and denoise. Enhance without Diffusion leaves out Regenerate-Q for a smaller download and faster builds. Refine installs the Refine tools: segmentation, metric and relative depth, and denoise. A preset downloads the weights and builds the engines in one step with one overall progress bar; you can keep using the app while it runs. Installs that came with the Essentials bundle already satisfy every preset.
Regenerate-Q tile size
Regenerate-Q processes a frame either whole or as a grid of overlapping tiles. The tile size is chosen per job: Enhance shows the choice as Tile size · Auto in the Enhancement card, with a note such as whole frame or 4 tiles. Pick another size from the list to override it for one job. Smaller tiles need less memory and run faster per tile, but each one sees less of the frame, so Auto never suggests tiles shorter than Settings → Application → Smallest suggested tile size (512 px by default; a pinned or Custom size may go down to 256 px).
On Apple Silicon the sizes that matter are the block-sparse ones: a tile that covers the whole output in one piece (1152×1920 for 1080p, 768×1280 for 720p, 640×1024 or 640×640 for SD) when the memory budget allows, otherwise the split that runs fastest. Auto can also pick a size that is not installed yet: the engine builds it on first use (a few seconds, plus a one-time compile on the first job), and the Model Library lists it with an Auto-built tag. The Essentials bundles ship the 640×640 and 640×1024 sizes on the 16 GB edition and add 768×1280 and 1152×1920 on the 32 GB edition; every size shares one set of Regenerate-Q weights, which the bundle also carries.
Each size keeps a compiled copy for the Neural Engine and GPU, which is what the CoreML cache line counts. Settings → Application → CoreML cache limit keeps those caches under a size you choose (20 GB by default) by removing the least recently used tile sizes; installed sizes stay and are recompiled when next used. Pick Custom… in the Tile size menu to enter any size on a 128 px grid.
Building engines for new dimensions
Engines are compiled per resolution. If you load a video in the Enhance workflow and no engine matches its shape, build one: open the family, click Build engine, and set the dimensions. The drawer validates as you type (dimensions must divide evenly, and it suggests the nearest valid values) and shows a live log while building. Builds take a few minutes per engine.
Build options
| Tile height / width | The input tile the engine processes at once. For Refocus, output is 4× the tile. Match your footage: 270×480 for vertical phone video, 480×270 for landscape. |
| Min / Max frames | The temporal window the engine accepts: how many frames it batches per inference. |
| Block-sparse attention (Regenerate-Q) | On by default. Builds a whole-frame tile (1152×1920 by default) that runs the transformer on the GPU with block-sparse attention; height and width must be multiples of 128. Off builds a dense tile in the older style. |
| Quantize (Regenerate) | int8 and palettized variants trade a small amount of quality for smaller, faster engines. |
| Model size (Segmentation) | Tiny through Large. Larger models track objects more accurately but load and run slower. |
| Variant · Size (Depth) | Relative or Metric depth, in Small / Base / Large. Metric gives real distances in meters, which depth masks can use. |
| Overwrite toggle | Rebuild over an existing engine of the same configuration. |
Engine · Auto).Performance: build for your hardware
The Essentials bundle and the presets build one general-purpose shape per model. The largest speedups come from engines shaped for your footage and compiled for your exact hardware, and every build is a one-time step: weights stay installed, and each new engine appears next to the others in step 3.
Custom model shapes on Apple Silicon
CoreML engines are compiled for a fixed tile and frame window. A tile that matches your footage's aspect ratio covers a frame in fewer passes with fewer seams: 480×270 for landscape, 270×480 for vertical video. A larger tile, such as 540×960 for 1080p sources, halves the number of passes again at the cost of memory; a larger frame window (Max frames) keeps the GPU busier per inference. Build in the Model Library, then pick the engine in Enhance or leave Engine · Auto to choose the best fit.
| Refocus tile | Match the aspect ratio of your sources. Go larger when memory allows; 16 GB machines are best at 270×480, 32 GB and up can take 540×960. |
| Regenerate-Q tile | Block-sparse sizes cover the output in one tile when the memory budget allows: 1152×1920 for 1080p output on Macs with a large budget, 768×1280 for 720p, 640×1024 or 640×640 for SD. Auto prefers them, builds a missing one on first use, and falls back to any dense 512 or 768 tile you still have installed when nothing block-sparse fits. |
| Frame window | Raise Max frames on machines with headroom to batch more frames per inference. |
| Segmentation size | Small is the default. Tiny tracks faster; Base+ and Large track more accurately but load and run slower. |
| Depth size | Small is the default and runs single-pass. Base and Large give finer maps at several times the cost. |
Managing disk space
The rail footer shows the total space used. Engines can be deleted individually from step 3 (trash icon) and rebuilt later. Weights stay installed, so rebuilding an engine does not re-download anything. Tile sizes the engine built for a job are tagged Auto-built; the CoreML cache limit in Settings → Application removes the least recently used ones when the caches grow past it.
Model updates arrive through Settings → Application → Software Update on macOS; see the settings reference.