🚀 V2 Update: The "Remastered" Edition
Dispatch Style V2 is a massive overhaul of the original LoRA. I rebuilt the dataset from scratch, expanding from 35 to 100+ high-quality images, manually captioned to ensure maximum flexibility and fidelity.
While V1 was an experiment, V2 is the definitive version. It captures the signature semi-3D/comic aesthetic of the Dispatch universe but renders it with higher fidelity, cleaner details, and better lighting than the game itself.
✨ Key Improvements in V2:
"Remastered" Look: The style is sharper, removing jagged edges and compression artifacts found in game screenshots. It looks like high-end concept art.
Generalization: Now works perfectly on Mechas/Robots, Scenery, Female Characters, and Animals (not just the main protagonist).
Smart Lighting: Natively understands
dramatic lightingandrim lightwithout needing complex prompts.No "Bleeding": Fixed overfitting issues where character details (like injuries) would appear randomly. Faces are now clean by default.
Depth of Field: The model now treats background blur as an artistic choice
depth of field), giving you crisp subjects every time.
🎮 How to use
Trigger Word:
dispatch_styleRecommended Strength:
0.9to1.0(V2 is very stable at full strength).Base Model: Highly recommended to use Illustrious Models for best results with the "western animation" aesthetic.
📝 Helper Tags (Optional but recommended):
Add this keyword to your prompt:
dispatch_style, western american animation, cel shading, gritty, depth of field, cinematic lighting,> western american animation, cel shading, gritty, depth of field, cinematic lighting
🚫 Negative Prompt:
> 3d render, cgi, realistic (If you want a flatter comic look)
> sketch, rough (If you want the clean game look)
💡 What this LoRA is good at:
Semi-3D Comic Aesthetic: The perfect blend between 2D drawing and 3D rendering.
Mechas & Tech: Incredible detail on robots and metallic surfaces (matte finish).
Atmosphere: High-contrast shadows, neon lights, and dark urban environments.
Action: Dynamic poses with particle effects and debris.
⚠️ Limitations:
Photorealism: This model fights against realism. It wants to make things look like a stylized cartoon.
Specific Actors: It captures the vibe of the characters, not their exact likeness (unless you prompt heavily for it).
This LoRA is a labor of love for the unique visual language of Dispatch. I hope it helps you create amazing stories in this gritty universe!
Description
First release of the Dispatch Style LoRA.
This version focuses on capturing the semi-3D comic aesthetic, gritty atmosphere and dramatic lighting of the Dispatch universe.
Trained on a curated dataset of 35 images with emphasis on consistent art direction, dark palette and cinematic shading.
FAQ
Comments (10)
On-site generation not enabled?
i activated it! I thought I had enabled it earlier, sorry
@SrMorningstar007 OK. Made and posted a picture (should be live in an hour). Couldn't really get it to apply heavily (I recommend a lot more pics than 35 for a general purpose style. 100+ something distinct pics with a variety of subjects seems to be the sweet spot. Even pics with no humans, just objects or backgrounds, work fine for getting this high.)
@NanashiAnon thank you very much for the tip. I used to make a lot of Loras before, but only of characters; this is my first one in a specific style, so I wanted to start with a few. I saw that it didn't turn out the way I wanted, so I'm already preparing for V2 to use more images...
@NanashiAnon If you want to achieve results like mine, use the tags + the WAI-illustrious-SDXL model. For now, this is the best I've been able to achieve with this LoRa.
@SrMorningstar007 I note with flip augmentation 35 should be doable for a narrow style (which is basically things like "particular video game's portraits", or "single character art on white background" that are only intended to make a very limited type of image), though more is better if you've got it
@NanashiAnon thank you again for the explanation. That makes a lot more sense now. Since this style is a bit broader than simple character portraits, I can see why adding more images would help the LoRA generalize better.
For the next version I will increase the dataset a lot and adjust the training steps accordingly. My plan is to use at least 85 steps per image, so the LoRA can learn the style more strongly and apply it more consistently across different types of scenes. I really appreciate your feedback. It is helping me improve V2 already.
hi, I released V2 today, and it's fantastic! I hope you enjoy it in your tests... 💫
Awesome style!
thank you very much! Stay tuned for the next update, which will bring many improvements...
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.






