Skip to content

Add SSV2 models & collection #498

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
joeyballentine merged 1 commit into from
Feb 19, 2026
Merged

Add SSV2 models & collection #498

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions data/collections.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,6 +72,18 @@
],
"author": "kim2091"
},
"c-starsample-v2-0": {
"name": "StarSample V2.0",
"author": "derpy",
"description": "**A collection of StarSample V2.0 models, now trained in a few architectures, rather than just one.**\n\nThe dataset for V2 is made completely from scratch (in relation to V1, but also in general), and includes what I'd consider many smaller datasets purpose-made to target areas where the models couldn't quite get up to the quality I wanted (particularly with focus on the heavier-weight models — the lighter architectures aren't really big enough for much effect). The improvement in dataset mostly, but also training heavier models has lead to a very large improvement, particularly with the \"HQ\" model.\n\nI wouldn't suggest using V1 over V2 in any circumstances — which I am very pleased to be able to say, having gone through at least 20 attempts to actually achieve that... somehow.",
"models": [
"2x-StarSample-V2-HQ",
"2x-StarSample-V2",
"2x-StarSample-V2-Lite",
"1x-StarSample-V2-NS",
"1x-StarSample-V2-Lite-NS"
]
},
"c-where-on-earth-is-carmen-sandiego-": {
"name": "Where On Earth Is Carmen Sandiego?",
"author": "pokepress",
Expand Down
100 changes: 100 additions & 0 deletions data/models/1x-StarSample-V2-Lite-NS.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,100 @@
{
"name": "StarSample V2.0 Lite NS",
"author": "derpy",
"license": "CC-BY-NC-SA-4.0",
"tags": [
"cartoon",
"compression-removal",
"general-upscaler",
"restoration"
],
"description": "This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on similar art.\n\nV2.0 greatly improves upon V1.0's dataset in every way, taking models from (realistically) only being viable at 1x, to now being far more competent at 2x, more so for the models trained with heavier architectures in this release. \n\nImprovements come as a significantly better understanding of compressions, and partly architecturally/partly dataset improved handling of details and overall understanding of content, leading to less artifacting and \"AI smudging\". The dataset takes from a larger variety of sources, despite being smaller than V1.0 (when tiled V1.0 would be 71,876 pairs), due to being filtered for IQA scores and detail density. It also contains many thousands of image pairs manually created to cover areas where there wasn't sufficient information.\n\nThis release also includes \"NS\", or \"No Scale\" models, which are a better representation of my initial goal with StarSample, and (StarSample V2.0 NS) should provide great 1x restoration results with little apparent artifacting, even where the heavier 2x models can fail due to having to increase resolution.\n\n- 2x StarSample V2.0 HQ — _(HAT-L)_\n- 2x StarSample V2.0 — _(ESRGAN)_\n- 2x StarSample V2.0 Lite — _(SPAN-S)_\n- 1x StarSample V2.0 NS — _(ESRGAN)_\n- 1x StarSample V2.0 Lite NS — _(SPAN-S)_ — **THIS MODEL**\n\n[Github Release](https://github.com/Derpiesaurus/models/releases/tag/v2.0_Lite_NS)",
"date": "2026-02-12",
"architecture": "span",
"size": [
"48nf"
],
"scale": 1,
"inputChannels": 3,
"outputChannels": 3,
"resources": [
{
"platform": "pytorch",
"type": "safetensors",
"size": 8888072,
"sha256": "5b5ee217b71793742ef11ce14c2114c6acae51ca92a2714a9770963c630046d2",
"urls": [
"https://github.com/Derpiesaurus/models/releases/download/v2.0_Lite_NS/1x_StarSample_V2.0_Lite_NS.safetensors"
]
},
{
"platform": "onnx",
"type": "onnx",
"size": 825471,
"sha256": "01e07b4a1f8ab9a27c163dbdc74a42a1f9760085ba3f1282ab1058af606ff321",
"urls": [
"https://github.com/Derpiesaurus/models/releases/download/v2.0_Lite_NS/1x_StarSample_V2.0_Lite_NS-fp16-opset17.onnx"
]
},
{
"platform": "onnx",
"type": "onnx",
"size": 1639156,
"sha256": "f91f29c70d46a85ea886d97f2e504867845f25f556d2e544bd2dceae1411db08",
"urls": [
"https://github.com/Derpiesaurus/models/releases/download/v2.0_Lite_NS/1x_StarSample_V2.0_Lite_NS-fp32-opset17.onnx"
]
}
],
"trainingIterations": 500000,
"trainingEpochs": 149,
"trainingBatchSize": 16,
"trainingHRSize": 128,
"trainingOTF": false,
"dataset": "HR = 4K GT uncompressed MLP: FiM episode frames + relevant uncompressed HR pairs to LR datasets /// LR = 1080p MLP: FiM episode frames sourced from YouTube in 3 different bitrates + custom MLP: FiM focal blur dataset + custom MLP: FiM GIF compression dataset in 3 different compression levels + custom MLP: FiM difficult details and other edge cases dataset + custom artificially-degraded MLP: FiM background dataset",
"datasetSize": 53560,
"images": [
{
"type": "paired",
"caption": "1 — 1x Example",
"LR": "https://i.slow.pics/H0fMm2gZ.webp",
"SR": "https://i.slow.pics/DITa8LRG.webp"
},
{
"type": "paired",
"caption": "2 — 1x Example",
"LR": "https://i.slow.pics/k371Z2k1.webp",
"SR": "https://i.slow.pics/0iDKrCCI.webp"
},
{
"type": "paired",
"caption": "3 — 1x Example",
"LR": "https://i.slow.pics/lcv8kGOM.webp",
"SR": "https://i.slow.pics/UHeXIqD7.webp"
},
{
"type": "paired",
"caption": "4 — 1x Example",
"LR": "https://i.slow.pics/JJrg8HrC.webp",
"SR": "https://i.slow.pics/JzOAd1oz.webp"
},
{
"type": "paired",
"caption": "5 — 1x Example",
"LR": "https://i.slow.pics/vsokWnr6.webp",
"SR": "https://i.slow.pics/iQqO2VWE.webp"
},
{
"type": "paired",
"caption": "6 — 1x Example",
"LR": "https://i.slow.pics/8di3VNuS.webp",
"SR": "https://i.slow.pics/Y8PTyN5h.webp"
},
{
"type": "paired",
"caption": "7 — 1x Example",
"LR": "https://i.slow.pics/bkvL11X3.webp",
"SR": "https://i.slow.pics/D8DTX1Yn.webp"
}
]
}
102 changes: 102 additions & 0 deletions data/models/1x-StarSample-V2-NS.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,102 @@
{
"name": "StarSample V2.0 NS",
"author": "derpy",
"license": "CC-BY-NC-SA-4.0",
"tags": [
"cartoon",
"compression-removal",
"general-upscaler",
"restoration"
],
"description": "This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on similar art.\n\nV2.0 greatly improves upon V1.0's dataset in every way, taking models from (realistically) only being viable at 1x, to now being far more competent at 2x, more so for the models trained with heavier architectures in this release. \n\nImprovements come as a significantly better understanding of compressions, and partly architecturally/partly dataset improved handling of details and overall understanding of content, leading to less artifacting and \"AI smudging\". The dataset takes from a larger variety of sources, despite being smaller than V1.0 (when tiled V1.0 would be 71,876 pairs), due to being filtered for IQA scores and detail density. It also contains many thousands of image pairs manually created to cover areas where there wasn't sufficient information.\n\nThis release also includes \"NS\", or \"No Scale\" models, which are a better representation of my initial goal with StarSample, and (StarSample V2.0 NS) should provide great 1x restoration results with little apparent artifacting, even where the heavier 2x models can fail due to having to increase resolution.\n\n- 2x StarSample V2.0 HQ — _(HAT-L)_\n- 2x StarSample V2.0 — _(ESRGAN)_\n- 2x StarSample V2.0 Lite — _(SPAN-S)_\n- 1x StarSample V2.0 NS — _(ESRGAN)_ — **THIS MODEL**\n- 1x StarSample V2.0 Lite NS — _(SPAN-S)_\n\n[Github Release](https://github.com/Derpiesaurus/models/releases/tag/v2.0_NS)",
"date": "2026-02-12",
"architecture": "esrgan",
"size": [
"64nf",
"23nb"
],
"scale": 1,
"inputChannels": 3,
"outputChannels": 3,
"resources": [
{
"platform": "pytorch",
"type": "safetensors",
"size": 66568516,
"sha256": "cc6dc5daf3eca585c359c0befb002996d389fef174e5de83a2f15ed74804d0d0",
"urls": [
"https://github.com/Derpiesaurus/models/releases/download/v2.0_NS/1x_StarSample_V2.0_NS.safetensors"
]
},
{
"platform": "onnx",
"type": "onnx",
"size": 33456842,
"sha256": "06f692cbe0860dbf1e81f7d8ea383e8ee1b2b96a12d75cd8736032bf1b4f77f0",
"urls": [
"https://github.com/Derpiesaurus/models/releases/download/v2.0_NS/1x_StarSample_V2.0_NS-fp16-opset17.onnx"
]
},
{
"platform": "onnx",
"type": "onnx",
"size": 66706120,
"sha256": "02ad068ebf4d82e49ef524a312be63dd19a37369bf64fae7e11c39bd8a7192d3",
"urls": [
"https://github.com/Derpiesaurus/models/releases/download/v2.0_NS/1x_StarSample_V2.0_NS-fp32-opset17.onnx"
]
}
],
"trainingIterations": 480000,
"trainingEpochs": 71,
"trainingBatchSize": 8,
"trainingHRSize": 192,
"trainingOTF": false,
"dataset": "HR = 4K GT uncompressed MLP: FiM episode frames + relevant uncompressed HR pairs to LR datasets /// LR = 1080p MLP: FiM episode frames sourced from YouTube in 3 different bitrates + custom MLP: FiM focal blur dataset + custom MLP: FiM GIF compression dataset in 3 different compression levels + custom MLP: FiM difficult details and other edge cases dataset + custom artificially-degraded MLP: FiM background dataset",
"datasetSize": 53560,
"pretrainedModelG": "1x-ESRGAN",
"images": [
{
"type": "paired",
"caption": "1 — 1x Example",
"LR": "https://i.slow.pics/HS5h28ub.webp",
"SR": "https://i.slow.pics/H74gbNDX.webp"
},
{
"type": "paired",
"caption": "2 — 1x Example",
"LR": "https://i.slow.pics/oqwzckjJ.webp",
"SR": "https://i.slow.pics/4eSEFaip.webp"
},
{
"type": "paired",
"caption": "3 — 1x Example",
"LR": "https://i.slow.pics/8kNqq0HT.webp",
"SR": "https://i.slow.pics/1oatlWpE.webp"
},
{
"type": "paired",
"caption": "4 — 1x Example",
"LR": "https://i.slow.pics/84LhArVj.webp",
"SR": "https://i.slow.pics/QcMfIfi5.webp"
},
{
"type": "paired",
"caption": "5 — 1x Example",
"LR": "https://i.slow.pics/AjQaq6xO.webp",
"SR": "https://i.slow.pics/BRpQVH0j.webp"
},
{
"type": "paired",
"caption": "6 — 1x Example",
"LR": "https://i.slow.pics/iLaE4j8r.webp",
"SR": "https://i.slow.pics/zZU3h8L0.webp"
},
{
"type": "paired",
"caption": "7 — 1x Example",
"LR": "https://i.slow.pics/Ud61KEgm.webp",
"SR": "https://i.slow.pics/FMtrl3Re.webp"
}
]
}
8 changes: 2 additions & 6 deletions data/models/2x-MLP-StarSample-V1-0.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,16 +1,12 @@
{
"name": "MLP StarSample V1.0",
"name": "StarSample V1.0",
"author": "derpy",
"license": "CC0-1.0",
"tags": [
"cartoon",
"colorization",
"compression-removal",
"deblur",
"denoise",
"general-upscaler",
"restoration",
"texture-generation"
"restoration"
],
"description": "This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on all similar art. \n\nIt was trained in 2x on ground truth 3840x2160 HRs and 1920x1080 LRs of varying compression, so it is able to upscale from 1080p to 2160p, where its detail retention is great, however it may create noticeable artifacting if looked at closely, like areas of randomly coloured pixels along edges. In 1x or 1.5x (2x upscaled and then downscaled back down) it performs extremely well, almost perfectly in fact, in correcting colours, removing compression, and crisping up lines - and this is the way the model is intended to be used (hence the acronym of its name being \"SS\", or \"supersampling\").\n\n**[Github Release](https://github.com/Derpiesaurus/models/releases/tag/v1.0)** \n**Showcase:** https://slow.pics/s/1ixqCSjy",
"date": "2024-05-22",
Expand Down
92 changes: 92 additions & 0 deletions data/models/2x-StarSample-V2-HQ.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,92 @@
{
"name": "StarSample V2.0 HQ",
"author": "derpy",
"license": "CC-BY-NC-SA-4.0",
"tags": [
"cartoon",
"compression-removal",
"general-upscaler",
"restoration"
],
"description": "This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on similar art.\n\nV2.0 greatly improves upon V1.0's dataset in every way, taking models from (realistically) only being viable at 1x, to now being far more competent at 2x, more so for the models trained with heavier architectures in this release. \n\nImprovements come as a significantly better understanding of compressions, and partly architecturally/partly dataset improved handling of details and overall understanding of content, leading to less artifacting and \"AI smudging\". The dataset takes from a larger variety of sources, despite being smaller than V1.0 (when tiled V1.0 would be 71,876 pairs), due to being filtered for IQA scores and detail density. It also contains many thousands of image pairs manually created to cover areas where there wasn't sufficient information.\n\nThis release also includes \"NS\", or \"No Scale\" models, which are a better representation of my initial goal with StarSample, and (StarSample V2.0 NS) should provide great 1x restoration results with little apparent artifacting, even where the heavier 2x models can fail due to having to increase resolution.\n\n- 2x StarSample V2.0 HQ — _(HAT-L)_ — **THIS MODEL**\n- 2x StarSample V2.0 — _(ESRGAN)_\n- 2x StarSample V2.0 Lite — _(SPAN-S)_\n- 1x StarSample V2.0 NS — _(ESRGAN)_\n- 1x StarSample V2.0 Lite NS — _(SPAN-S)_\n\n[Github Release](https://github.com/Derpiesaurus/models/releases/tag/v2.0_HQ)",
"date": "2026-02-12",
"architecture": "hat",
"size": [
"64nf",
"180dim"
],
"scale": 2,
"inputChannels": 3,
"outputChannels": 3,
"resources": [
{
"platform": "pytorch",
"type": "safetensors",
"size": 164708980,
"sha256": "77115df5bb80600a32a762e321e26f2a0114142d00ed060a7e88a5be269d5b05",
"urls": [
"https://github.com/Derpiesaurus/models/releases/download/v2.0_HQ/2x_StarSample_V2.0_HQ.safetensors"
]
},
{
"platform": "onnx",
"type": "onnx",
"size": 318442478,
"sha256": "b10cb249b75d55ed6ad0800ef825e750940582a72e219a2b2b781789381615e3",
"urls": [
"https://github.com/Derpiesaurus/models/releases/download/v2.0_HQ/2x_StarSample_V2.0_HQ-fp32-opset17.onnx"
]
}
],
"trainingIterations": 235000,
"trainingEpochs": 17,
"trainingBatchSize": 4,
"trainingHRSize": 256,
"trainingOTF": false,
"dataset": "HR = 4K GT uncompressed MLP: FiM episode frames + relevant uncompressed HR pairs to LR datasets /// LR = 1080p MLP: FiM episode frames sourced from YouTube in 3 different bitrates + custom MLP: FiM focal blur dataset + custom MLP: FiM GIF compression dataset in 3 different compression levels + custom MLP: FiM difficult details and other edge cases dataset + custom artificially-degraded MLP: FiM background dataset",
"datasetSize": 53560,
"images": [
{
"type": "paired",
"caption": "1 — 2x Example",
"LR": "https://i.slow.pics/IKXCPxRV.webp",
"SR": "https://i.slow.pics/cHfd9R4G.webp"
},
{
"type": "paired",
"caption": "2 — 2x Example",
"LR": "https://i.slow.pics/rlNFQRvB.webp",
"SR": "https://i.slow.pics/h2Bn7JqX.webp"
},
{
"type": "paired",
"caption": "3 — 2x Example",
"LR": "https://i.slow.pics/dg1MZfbi.webp",
"SR": "https://i.slow.pics/W0FnFNHq.webp"
},
{
"type": "paired",
"caption": "4 — 2x Example",
"LR": "https://i.slow.pics/1UGqZyce.webp",
"SR": "https://i.slow.pics/egGU6NOO.webp"
},
{
"type": "paired",
"caption": "5 — 2x Example",
"LR": "https://i.slow.pics/18h3dSQA.webp",
"SR": "https://i.slow.pics/PGDlM6Zr.webp"
},
{
"type": "paired",
"caption": "6 — 2x Example",
"LR": "https://i.slow.pics/uVcHz2Qh.webp",
"SR": "https://i.slow.pics/ZyjAACpb.webp"
},
{
"type": "paired",
"caption": "7 — 2x Example",
"LR": "https://i.slow.pics/Agz0NTq2.webp",
"SR": "https://i.slow.pics/vYuzS96I.webp"
}
]
}
Loading