@Dr. Furkan Gözükara For fine-tuning flux with kohya, do you recommend using the models from OwlMast

Prompt: A close-up shot of a smiling boy with a joyful expression, sitting comfortably in a cozy room, surrounded by soft toys and books. The boy has curly hair and wears a colorful t-shirt. Bright, warm lighting highlights his happy face. Close-up view.
Negative Prompt: Overexposure, static, blurred details, subtitles, paintings, pictures, still, overall gray, worst quality, low quality, JPEG compression residue, ugly, mutilated, redundant fingers, poorly painted hands, poorly painted faces, deformed, disfigured, deformed limbs, fused fingers, cluttered background, three legs, a lot of people in the background, upside down
Used Model: WAN 2.1 14B Text-to-Video
Number of Inference Steps: 50
Seed: 420498510
Number of Frames: 81
Denoising Strength: N/A
LoRA Model: None
TeaCache Enabled: True
TeaCache L1 Threshold: 0.15
TeaCache Model ID: Wan2.1-T2V-14B
Precision: BF16
Auto Crop: Enabled
Final Resolution: 832x480
Generation Duration: 320.97 seconds

Prompt: A close-up shot of a smiling boy with a joyful expression, sitting comfortably in a cozy room, surrounded by soft toys and books. The boy has curly hair and wears a colorful t-shirt. Bright, warm lighting highlights his happy face. Close-up view.
Negative Prompt: Overexposure, static, blurred details, subtitles, paintings, pictures, still, overall gray, worst quality, low quality, JPEG compression residue, ugly, mutilated, redundant fingers, poorly painted hands, poorly painted faces, deformed, disfigured, deformed limbs, fused fingers, cluttered background, three legs, a lot of people in the background, upside down
Used Model: WAN 2.1 14B Text-to-Video
Number of Inference Steps: 50
Seed: 420498510
Number of Frames: 81
Denoising Strength: N/A
LoRA Model: None
TeaCache Enabled: True
TeaCache L1 Threshold: 0.15
TeaCache Model ID: Wan2.1-T2V-14B
Precision: BF16
Auto Crop: Enabled
Final Resolution: 832x480
Generation Duration: 320.97 seconds

Prompt: A close-up shot of a smiling boy with a joyful expression, sitting comfortably in a cozy room, surrounded by soft toys and books. The boy has curly hair and wears a colorful t-shirt. Bright, warm lighting highlights his happy face. Close-up view.
Negative Prompt: Overexposure, static, blurred details, subtitles, paintings, pictures, still, overall gray, worst quality, low quality, JPEG compression residue, ugly, mutilated, redundant fingers, poorly painted hands, poorly painted faces, deformed, disfigured, deformed limbs, fused fingers, cluttered background, three legs, a lot of people in the background, upside down
Used Model: WAN 2.1 14B Text-to-Video
Number of Inference Steps: 50
Seed: 420498510
Number of Frames: 81
Denoising Strength: N/A
LoRA Model: None
TeaCache Enabled: True
TeaCache L1 Threshold: 0.15
TeaCache Model ID: Wan2.1-T2V-14B
Precision: BF16
Auto Crop: Enabled
Final Resolution: 832x480
Generation Duration: 320.97 seconds

Prompt: A close-up shot of a smiling boy with a joyful expression, sitting comfortably in a cozy room, surrounded by soft toys and books. The boy has curly hair and wears a colorful t-shirt. Bright, warm lighting highlights his happy face. Close-up view.
Negative Prompt: Overexposure, static, blurred details, subtitles, paintings, pictures, still, overall gray, worst quality, low quality, JPEG compression residue, ugly, mutilated, redundant fingers, poorly painted hands, poorly painted faces, deformed, disfigured, deformed limbs, fused fingers, cluttered background, three legs, a lot of people in the background, upside down
Used Model: WAN 2.1 14B Text-to-Video
Number of Inference Steps: 50
Seed: 420498510
Number of Frames: 81
Denoising Strength: N/A
LoRA Model: None
TeaCache Enabled: True
TeaCache L1 Threshold: 0.15
TeaCache Model ID: Wan2.1-T2V-14B
Precision: BF16
Auto Crop: Enabled
Final Resolution: 832x480
Generation Duration: 320.97 seconds

FFurkan Gözükara SECourses ```Prompt: A close-up shot of a smiling boy with a joyful expression, sitting co...

A

AiInfluence•3/15/25, 1:57 PM

How should we value TeaCache Threshold? Like what are max min values and. what to consider for when adjusting

AAiInfluence How should we value TeaCache Threshold? Like what are max min values and. what t...

F

Furkan Gözükara SECourses•3/15/25, 1:57 PM

0.15 is pretty much max i believe. above will make faster

F

Furkan Gözükara SECourses•3/15/25, 1:57 PM

but quality may degrade

F

Furkan Gözükara SECourses•3/15/25, 1:57 PM

try 0.2

F

Furkan Gözükara SECourses•3/15/25, 1:57 PM

over 100% speed probably

FFurkan Gözükara SECourses 0.15 is pretty much max i believe. above will make faster

A

AiInfluence•3/15/25, 2:03 PM

and for gpu L40?

AAiInfluence and for gpu L40?

F

Furkan Gözükara SECourses•3/15/25, 2:09 PM

same you can try to load all params in gpu

F

Furkan Gözükara SECourses•3/15/25, 2:09 PM

if works

F

Furkan Gözükara SECourses•3/15/25, 2:09 PM

someone said there is a banding problem with RIFE

F

Furkan Gözükara SECourses•3/15/25, 2:09 PM

is he here?

J

JaviEM•3/15/25, 3:12 PM

Can the RAM bug cause quality degradation? I've made about 3 or 4 videos in a row with awful results, each one worse than the previous one. I restarted the application and it generated a much more natural video.

JJaviEM Can the RAM bug cause quality degradation? I've made about 3 or 4 videos in a ro...

F

Furkan Gözükara SECourses•3/15/25, 3:40 PM

i doubt that. actually i will now try several consquent ones lets see what happens

F

Furkan Gözükara SECourses•3/15/25, 3:43 PM

testing multi line

F

Furkan Gözükara SECourses•3/15/25, 3:43 PM

J

JaviEM•3/15/25, 4:00 PM

in my experience long promps have always given me bad results

F

Furkan Gözükara SECourses•3/15/25, 4:43 PM

@JaviEM all looking normal

F

Furkan Gözükara SECourses•3/15/25, 4:43 PM

you doing image to video?

F

Furkan Gözükara SECourses•3/15/25, 4:44 PM

F

Furkan Gözükara SECourses•3/15/25, 4:44 PM

00152.txt978B

00154.txt982B

00153.txt978B

FFurkan Gözükara SECourses you doing image to video?

J

JaviEM•3/15/25, 4:50 PM

Yes. I am getting very mixed results. Some videos are perfect, some are awful.

J

JaviEM•3/15/25, 4:51 PM

It's not funny to wait 1 hour for a horrible video with an extreme hallucination

J

JaviEM•3/15/25, 4:52 PM

Would it be possible to include some kind of preview while it is being processed? even if it is just a random image of the video to see if everything is correct or if it is better to cancel it.

JJaviEM Would it be possible to include some kind of preview while it is being processed...

F

Furkan Gözükara SECourses•3/15/25, 4:53 PM

i am not sure since vae decoding happens at the end.

F

Furkan Gözükara SECourses•3/15/25, 4:54 PM

lets say you generate exactly same video

@Dr. Furkan Gözükara For fine-tuning flux with kohya, do you recommend using the models from OwlMast

Similar Threads

Similar Threads

Similar Threads