Hugging Face's logo

yijunwang2
/
krea2-outpaint

Image-to-Image
Diffusers
krea-2
lora
outpainting
image-editing
Model card Files
xet
 Community
3
yijunwang2 commited on
Commit
9b1c873
·
verified ·
1 Parent(s): 1017070

Upload private Krea 2 Outpaint release candidate

Browse files
Files changed (2) hide show
  1. README.md +4 -6
  2. SHA256SUMS +1 -1
README.md CHANGED
@@ -3,6 +3,7 @@ license: other
3
  license_name: krea-2-community-license
4
  license_link: https://krea.ai/krea-2-licensing
5
  base_model: krea/Krea-2-Turbo
 
6
  pipeline_tag: image-to-image
7
  tags:
8
  - krea-2
@@ -14,9 +15,6 @@ tags:
14
 
15
  # Krea 2 Registered Outpaint
16
 
17
- > Draft release candidate. This repository is private while its documentation,
18
- > license files, and clean-download inference path are reviewed.
19
-
20
  This rank-32 LoRA extends an image into a larger canvas while preserving its
21
  known pixels and registering the source image at an explicit canvas location.
22
  It was trained against Krea 2 Raw and is intended for distilled 8-step inference
@@ -45,7 +43,7 @@ showcase/ fully synthetic source/output examples
45
 
46
  ## Tested capabilities
47
 
48
- The release candidate has been tested across photorealistic, watercolor, and
49
  stylized 3D scenes, with the known source placed on different parts of the
50
  canvas:
51
 
@@ -159,8 +157,8 @@ small decode differences at the boundary.
159
  extended.
160
  - The 384-pixel conditioning budget prioritizes speed and structure over exact
161
  recovery of tiny source details; final compositing preserves the known pixels.
162
- - BF16 clean-download inference remains a release-blocking verification item for
163
- this draft.
164
 
165
  ## Training summary
166
 
 
3
  license_name: krea-2-community-license
4
  license_link: https://krea.ai/krea-2-licensing
5
  base_model: krea/Krea-2-Turbo
6
+ library_name: diffusers
7
  pipeline_tag: image-to-image
8
  tags:
9
  - krea-2
 
15
 
16
  # Krea 2 Registered Outpaint
17
 
 
 
 
18
  This rank-32 LoRA extends an image into a larger canvas while preserving its
19
  known pixels and registering the source image at an explicit canvas location.
20
  It was trained against Krea 2 Raw and is intended for distilled 8-step inference
 
43
 
44
  ## Tested capabilities
45
 
46
+ This release has been tested across photorealistic, watercolor, and
47
  stylized 3D scenes, with the known source placed on different parts of the
48
  canvas:
49
 
 
157
  extended.
158
  - The 384-pixel conditioning budget prioritizes speed and structure over exact
159
  recovery of tiny source details; final compositing preserves the known pixels.
160
+ - The included BF16 example is provided as a portable reference path; the
161
+ reported performance measurements below are from the tested INT8 runtime.
162
 
163
  ## Training summary
164
 
SHA256SUMS CHANGED
@@ -1,7 +1,7 @@
1
  b82a2805162bde714a4eb27b9063c4fc3345d08a30be055134a6160e5430ba74 LICENSE.pdf
2
  7b011fa1e03a47d475c67bc42d808ba6b585b8e90b55dd856b8571fb6683c05d NOTICE
3
  cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30 PIPELINE_LICENSE
4
- 17abbccae531319bc804f1d672115afc77959198f9bb4402480f68eb8a166b8c README.md
5
  b040b84b01857e953e551e84502591c9bd828b67f89c6c9b6e048484a02b3a82 example.py
6
  1de7c106ae978de21433152bdb4ba6b09118b33be81b7ee4851420f0013a7e76 krea2_outpaint_rank32.safetensors
7
  fc176da8a96ee0d62651be5263f18a0bc8b1f9cd243111a31ddce05a62851024 outpaint.py
 
1
  b82a2805162bde714a4eb27b9063c4fc3345d08a30be055134a6160e5430ba74 LICENSE.pdf
2
  7b011fa1e03a47d475c67bc42d808ba6b585b8e90b55dd856b8571fb6683c05d NOTICE
3
  cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30 PIPELINE_LICENSE
4
+ 9ec9ab03827d1b7a747f60f143f1577101ca641a301be37283c339d804645ff1 README.md
5
  b040b84b01857e953e551e84502591c9bd828b67f89c6c9b6e048484a02b3a82 example.py
6
  1de7c106ae978de21433152bdb4ba6b09118b33be81b7ee4851420f0013a7e76 krea2_outpaint_rank32.safetensors
7
  fc176da8a96ee0d62651be5263f18a0bc8b1f9cd243111a31ddce05a62851024 outpaint.py