ID-GEO (Generic Enhancement Operator)
Experimental Ideogram 4 research LoRA targeting the unconditional branch.
Developed to explore guidance-prior modification and its influence on generated images.
Designed primarily for img2img workflows, but compatible with txt2img as well.
Recommended starting strength: 0.5–0.7
Observed sweet spot: ~0.8
Examples
1. Photorealistic Action Portrait
Human, animal, and motion-detail preservation across increasing GEO strengths.
2. Sci-Fi Environment
Comparison of material rendering, lighting behavior, and structural consistency.
3. Close-Up Portrait
Evaluation of skin texture, hair separation, and facial identity stability.
4. Typography
BBox-controlled typography comparison focusing on glyph integrity, spacing, and embossed detail.
5. Anime Character
Anime-style comparison evaluating linework, cel-shading stability, and background signage.
Notes
Intended for use on the Ideogram 4 unconditional branch.
Experimental research project.
Results may vary depending on workflow, guidance settings, denoise strength, and additional LoRAs.
Description
v1
FAQ
Comments (6)
txt2img workflows included in most images, img2img workflow is in "Optional files"
Can you explain in simpler terms what this LoRA does? Thanks!
In simpler terms, this LoRA functions as a general quality-booster and subtle detail-enhancer for your AI generations, rather than generating a specific character, outfit, or art style.
("My gens are now look less AI")
in simple tempt it makes output a mess garbage. Even on 0.3 output become messed.
@velanteg You’ve got a massive mouth for someone who hasn’t posted a single shred of proof, a workflow, or an example to back it up. Classic troll.
@velanteg What an asinine and inane, if not imbecilic comment.