Sentinel7 commited on
Commit
627e3a9
·
verified ·
1 Parent(s): 07413fe

Upload 3008549/3416236/README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. 3008549/3416236/README.md +94 -0
3008549/3416236/README.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ tags:
4
+ - concept
5
+ ---
6
+
7
+ # Capitan01R's Krea2 TextFusion Refusal-Reduction - v2 - 3416236
8
+
9
+ **Model Type**: LORA
10
+
11
+ **Base Model**: Krea 2
12
+
13
+ **Trigger Words**: None
14
+
15
+ **Tags**: concept
16
+
17
+ ## Gallery
18
+
19
+ <table>
20
+ <tr>
21
+ <td><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a3532f3c-5e3f-45eb-ab0e-56a13b65bab6/original=true/146174399.jpeg" width="200"></td>
22
+ </tr>
23
+ </table>
24
+
25
+ ## Description
26
+
27
+ **Krea2 TextFusion Refusal-Reduction [by Capitan01R]**
28
+ ======================================================
29
+
30
+ No idea what happened with Capitan01R
31
+
32
+ I will happily take the model down if [Capitan01R](https://civitai.red/user/Capitan01R) decides to come back. Until then yellow buzz generated with this gets tipped to him. If he deletes his account I'll throw the buzz to some random dude.
33
+
34
+ But this is a great LoRA and it should be available to the community.
35
+
36
+ ---
37
+
38
+ # Krea2 TextFusion Refusal-Reduction LoRA
39
+
40
+ This is a dedicated rank-64 LoRA trained with one narrow objective: reduce Krea2’s learned refusal and restriction behavior while preserving as much of the base model’s existing visual knowledge and prompt behavior as possible.
41
+
42
+ This is not a concept, character, style, anatomy, or aesthetic LoRA. It was not trained to add new visual concepts, reproduce a training-image set, or impose a preferred look. Its training objective was exclusively to push the model away from refusal behavior so that concepts already represented within the base model are less likely to be suppressed during text conditioning.
43
+
44
+ ## What It Changes
45
+
46
+ Krea2 receives multiple hidden-state taps from its Qwen-VL text encoder and processes them through TextFusion before sending the resulting conditioning into the image transformer. This LoRA applies learned low-rank residuals only to the attention and internal MLP projections within:
47
+
48
+ - txtfusion.layerwise\_blocks.0
49
+
50
+ - txtfusion.layerwise\_blocks.1
51
+
52
+ - txtfusion.refiner\_blocks.0
53
+
54
+ - txtfusion.refiner\_blocks.1
55
+
56
+ The release version deliberately contains no adapter for:
57
+
58
+ - The TextFusion 1 × 12 tap projector
59
+
60
+ - The external/general txtmlp
61
+
62
+ - The image-transformer blocks or any other image-generation layers
63
+
64
+ ## How It Works
65
+
66
+ The base checkpoint is never overwritten. For every targeted linear layer, LoRA adds a learned rank-64 residual to the original weight during inference:
67
+
68
+ W\_effective = W\_base + strength × ΔW
69
+
70
+ This changes how existing Qwen-VL text features are routed, gated, transported, and refined through TextFusion before they reach Krea2’s untouched image transformer. In other words, the LoRA is intended to improve access to visual knowledge already present in the base model, not inject a new concept or replace the model’s learned visual representations.
71
+
72
+ This release targets the specific TextFusion route isolated through layer-by-layer ablation instead of broadly amplifying activations or directly altering the projector.
73
+
74
+ ## Usage
75
+
76
+ **Use strength of "1.00"**
77
+
78
+ **Responsibility and Output Disclaimer**
79
+
80
+ This LoRA modifies the model’s text-conditioning behavior but does not determine, authorize, endorse, supervise, or control the images produced by individual users, prompts, workflows, checkpoints, samplers, or third-party software.
81
+
82
+ Generated outputs remain dependent on the base model, user-provided conditioning, inference configuration, and the surrounding generation pipeline. The author assumes no responsibility or liability for what users choose to generate, how generated material is used, or whether any output complies with applicable laws, platform rules, licensing terms, or third-party rights.
83
+
84
+ Users are solely responsible for their prompts, generated outputs, distribution decisions, and use of this LoRA.
85
+
86
+ ---
87
+
88
+ Author: [CraigToomy](https://civitai.com/user/CraigToomy)
89
+
90
+ Model: [CivitAI Model Page](https://civitai.com/models/3008549?modelVersionId=3416236)
91
+
92
+ Archive: [CivArchive Page](https://civarchive.com/models/3008549?modelVersionId=3416236)
93
+
94
+ <!-- Version: 20261020_upload -->