❌💲 Commercial use of this model is not allowed

I was surprised that I couldn't find any Thomas and Friends LoRAs for base models other than SD1.5, I thought I'd give this a shot.

I never trained a LoRA for a vehicle before, and this case is made more interesting by the fact Thomas is a locomotive with a humanoid face. Please note I'm fairly new to this franchise and may not immediately understand train-related terms.

🖼DATASET

Mainly screenshots from some specials (1080p). Tagged manually.

Railway tracks are tagged, as they're not present in some images.

Some images contain other characters, mainly Thomas's crew.

Some images contain "unusual" states like being dirty, scratched up, covered in snow, wet or derailed. Datasets for versions past PonyXL V1 have some pictures where his livery isn't blue ("alternate livery color").

⚠ISSUES

NoobEps V1: base Noob results may look like hot garbage, try 3D finetunes instead; face may turn pink as if it was human

PDXL V2: In my base Pony tests I found it may produce pink-tinted images unprompted. There weren't any images like that in the dataset. I have no idea why this happens. This should be avoidable if you put "pink tint, pink theme" into negative prompt.

General deformities: wheels messed up / in the wrong amount, extra dome, lamp duplicating / on wrong side, etc. Colouring errors.

Don't try rear view please.

Wrinkles between eyebrows may be too much.

Bad railway tracks, although I don't expect a character-focused base model to do those well.

Description

- I felt that the number of images was quite low, and they were lacking in angles. It has been increased to 130. Most of the new ones are from The Adventure Begins and Journey Beyond Sodor, I got them from YouTube uploads so they may be a little crusty. I considered other specials but didn't find any more in 1080p. There's a gajillion of episodes, but I got too impatient to look into them (and frankly I don't like Thomas the character all that much). I included a few rear view shots, but I guess not enough because I get malformed results.

- Because I used pictures from TAB I included some where his livery is colored differently, those are tagged with "alternate livery color". They make up a VERY small portion of the dataset, so hopefully they don't cause unwanted side effects.

- Added "vehicle focus" to trigger words

- Replaced tag "number 1 on livery" with "number 1 on side" for consistency with another tag I added

- Images where his face is not visible do NOT use "male focus" and "animate inanimate"

- Minor tag edits