Purpose of this LoRA:

While I do love the other Fifi la Fume LoRAs, this one exists to express the niche Tokyo Movie Studio (TMS) design and depiction of the character. Every famous depiction of Fifi la Fume you can think of has been penned by TMS. It is her peak form and construction. Below is a side by side comparison.

[NEW] Illustrious v2.3:

After spending a hefty 5000 buzz on training, experimenting, and re-training over a span of 3 weeks, I now reveal a new better, stronger LoRA for more Fifi TMS goodness!

• Trigger is "fifitms"

• No need for the 20+ tags to replicate Fifi la Fume. The trigger is all you'll need.

Prompt Suggestions

• Because of how TMS forgets to draw the bow behind her ear, I suggest to add "bow ribbon" to ensure its generation if it bothers you

• Recommended LoRA Weight: 0.7-1.0

• Positive Prompt: fifitms, bow ribbon, short stack, fluffy tail, close-up

• Negative Prompt: eyeshadow

• Avoid tagging Fifi la Fume, skunk, purple fur, purple hair and similar tags. Remember, this LoRA only uses "perfect" TMS Fifi images. Even if your checkpoint is aware of the Fifi La Fume tag, including it will only muddle the output, not make it better.

• The Amazing Three Outfit: pink dress, red leggings, red belt, black high heels, blue eyeshadow

• Early Green Outfit Design: green shirt, green skirt

Updates over v1.b and v1.0

• Manual and changed tagging on every picture

• Removed weaker images

• Added angles tags such as three-quarter view, rear view

• Emotions and mouth shapes accurately tagged

Pony v1.b and v1.0:

Prompt Suggestions

• "Fifi la fume" is the trigger caption

• [[[skunk]]], [[[ear bow]]] for slightly more consistent results

• Recommended LoRA Weight: 0.6-0.8

• Pure TMS Style Weight: 1.0

  • Old School Style: "Screencap" (Credit to Thunderkiss89)

• Negative Prompt: "3D"

Strengths

• Very flexible with many poses and styles

• Less tagging is required for the LoRA to activate (only needing "Fifi la Fume" to trigger)

• Extremely toony and expressive!

• Can do SFW and NSFW interchangeably

Weaknesses

• It cannot do niche outfits that appear in one or two episodes of the show

• The short toony body sometimes has issues adhering to typical adult anthro bodies

• Occasionally, the tail is missing its signature stripe or is doubled

• Occasionally, TMS will forget to draw her bow. This LoRA will too

• TMS has slightly different depictions of Fifi. This LoRA will too

Future Updates:

At the moment, I can't think of any way to improve this LoRA. Her other outfits aren't drawn by TMS and would need to be stitched together to be included in the dataset. Maybe someone can point me to some directions.

Update 9-9-2025: There is a thought on if I should manually tweak every TMS image to be perfect. No coloring errors, re-added missing bows, cleaned up outlines. However if I did that, this LoRA wouldn't be TMS anymore, it would be TMS+Me Fifi LoRA. I believe we want authenticity, not so much the rose tinted version of authentic.

Misc Information:

This LoRA was trained using various snapshots from TMS eps and the "Tiny Toon Adventures: How I Spent My Vacation" movie. Snapshots were provided by @MalteserRefs from Twitter.

Illustrious: Special Thanks to @saransaran and @urusan for additional guidance on making a better dataset. Thanks to saransaran for his manual upscaling and touchup in the training data.

Pony: With extra help from the Furry Diffusion discord members on making the dataset better and proper tagging.