2x StarSample V2.0

StarSample V2.0

Scale: 2x
Architecture: ESRGAN

Author: .derpy.
License: CC BY-NC-SA
Purpose: Restoration, Compression Removal, General Upscaler
Subject: Cartoon
Input Type: Images
Date: February 12th 2026

Size: 64nf, 23nb
I/O Channels: 3(RGB)->3(RGB)

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.

Dataset Size: 53,560
OTF (on the fly augmentations): No
Pretrained Model: 2x-ESRGAN
Iterations: 500,000
Batch Size: 10
HR Size: 192
LR Size: 96

Description:
This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on similar art.

V2.0 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.

Improvements 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.

This 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.

Models (in order of quality):

• 2x StarSample V2.0 HQ — (HAT-L)
• 2x StarSample V2.0 — (ESRGAN) — THIS MODEL
• 2x StarSample V2.0 Lite — (SPAN-S)
• 1x StarSample V2.0 NS — (ESRGAN)
• 1x StarSample V2.0 Lite NS — (SPAN-S)

Showcase:

1x StarSample V2.0 NS

StarSample V2.0 No Scale

Scale: 1x
Architecture: ESRGAN

Author: .derpy.
License: CC BY-NC-SA
Purpose: Restoration, Compression Removal, General Upscaler
Subject: Cartoon
Input Type: Images
Date: February 12th 2026

Size: 64nf, 23nb
I/O Channels: 3(RGB)->3(RGB)

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.

Dataset Size: 53,560
OTF (on the fly augmentations): No
Pretrained Model: 1x-ESRGAN
Iterations: 480,000
Batch Size: 8
HR Size: 192
LR Size: 96

Description:
This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on similar art.

V2.0 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.

Improvements 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.

This 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.

Models (in order of quality):

• 2x StarSample V2.0 HQ — (HAT-L)
• 2x StarSample V2.0 — (ESRGAN)
• 2x StarSample V2.0 Lite — (SPAN-S)
• 1x StarSample V2.0 NS — (ESRGAN) — THIS MODEL
• 1x StarSample V2.0 Lite NS — (SPAN-S)

Showcase:

1x StarSample V2.0 Lite NS

StarSample V2.0 Lite No Scale

Scale: 1x
Architecture: SPAN-S

Author: .derpy.
License: CC BY-NC-SA
Purpose: Restoration, Compression Removal, General Upscaler
Subject: Cartoon
Input Type: Images
Date: February 12th 2026

Size: 48nf
I/O Channels: 3(RGB)->3(RGB)

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.

Dataset Size: 53,560
OTF (on the fly augmentations): No
Pretrained Model: 2x_BHI_small_Redux_SPAN_S_1m30k
Iterations: 500,000
Batch Size: 16
HR Size: 128
LR Size: 64

Description:
This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on similar art.

V2.0 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.

Improvements 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.

This 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.

Models (in order of quality):

• 2x StarSample V2.0 HQ — (HAT-L)
• 2x StarSample V2.0 — (ESRGAN)
• 2x StarSample V2.0 Lite — (SPAN-S)
• 1x StarSample V2.0 NS — (ESRGAN)
• 1x StarSample V2.0 Lite NS — (SPAN-S) — THIS MODEL

Showcase:

2x StarSample V2.0 Lite

StarSample V2.0 Lite

Scale: 2x
Architecture: SPAN-S

Author: .derpy.
License: CC BY-NC-SA
Purpose: Restoration, Compression Removal, General Upscaler
Subject: Cartoon
Input Type: Images
Date: February 12th 2026

Size: 48nf
I/O Channels: 3(RGB)->3(RGB)

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.

Dataset Size: 53,560
OTF (on the fly augmentations): No
Pretrained Model: 2x_BHI_small_Redux_SPAN_S_1m30k
Iterations: 500,000
Batch Size: 16
HR Size: 192
LR Size: 96

Description:
This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on similar art.

V2.0 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.

Improvements 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.

This 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.

Models (in order of quality):

• 2x StarSample V2.0 HQ — (HAT-L)
• 2x StarSample V2.0 — (ESRGAN)
• 2x StarSample V2.0 Lite — (SPAN-S) — THIS MODEL
• 1x StarSample V2.0 NS — (ESRGAN)
• 1x StarSample V2.0 Lite NS — (SPAN-S)

Showcase:

2x StarSample V2.0 HQ

StarSample V2.0 High Quality

Scale: 2x
Architecture: HAT-L

Author: .derpy.
License: CC BY-NC-SA
Purpose: Restoration, Compression Removal, General Upscaler
Subject: Cartoon
Input Type: Images
Date: February 12th 2026

Size: 64nf, 180dim
I/O Channels: 3(RGB)->3(RGB)

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.

Dataset Size: 53,560
OTF (on the fly augmentations): No
Pretrained Model: HAT-L_SRx2_ImageNet-pretrain
Iterations: 235,000
Batch Size: 4
HR Size: 256
LR Size: 128

Description:
This is a model for the restoration of My Little Pony: Friendship is Magic, however it also works decently well on similar art.

V2.0 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.

Improvements 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.

This 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.

Models (in order of quality):

• 2x StarSample V2.0 HQ — (HAT-L) — THIS MODEL
• 2x StarSample V2.0 — (ESRGAN)
• 2x StarSample V2.0 Lite — (SPAN-S)
• 1x StarSample V2.0 NS — (ESRGAN)
• 1x StarSample V2.0 Lite NS — (SPAN-S)

Showcase:

2x StarSample V1.0

StarSample V1.0

Scale: 2x (although trained in favour of 1x performance)
Architecture: SRVGGNetCompact

Author: .derpy.
License: CC0-1.0
Purpose: Restoration, Colour Correction, Compression Removal, General Upscaler
Subject: Cartoon
Input Type: Images
Date: May 22nd 2024

Size: 64nf, 16nc
I/O Channels: 3(RGB)->3(RGB)

Dataset: A mix of 4K GT show frames and background-only images.
Dataset Size: 2,567
OTF (on the fly augmentations): No
Pretrained Model: 2x_Compact_Pretrain
Iterations: 230,000
Batch Size: 28 (7 * 4 accumulate)
GT Size: 256

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.

It 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").

Showcase:
slow.pics Comparison