Stable diffusion thread

[doogyhatts]

HunyuanDiT is a bilingual Text to Image generation solution by Tencent.

Better than SD3?
The minimum GPU memory required is 11GB.

 

[zheng]

HunyuanDiT is a bilingual Text to Image generation solution by Tencent.
Better than SD3?

what is is not a sd3 model right?

 

[doogyhatts]

what is is not a sd3 model right?

Tried their demo. Prompt adherence looks good, just need integration into ComfyUI so can use controlnet.
Image quality looks different from SDXL.

Spoiler

 

[doogyhatts]

 

[doogyhatts]

Prompt:
1950's panavision 70 color film stock, beautiful cinematography, a giant white rabbit with pink eyes and two long ears, wears a red and white checkered coat, standing in the daisy flower field with surrounding plants and trees.

Results:
Not bad. It is able to get the basic analog film aesthetic.

 

[doogyhatts]

Prompt:
Audrey Hepburn holding an iPhone, 1950's aesthetic, made in the 1950's, super Panavision 70 color film stock, beautiful cinematography, low contrast, balanced colors, 3 point studio lighting, high resolution, rich color palette, smooth tonality

Spoiler

 

[RyanJ]

HunyuanDiT is a bilingual Text to Image generation solution by Tencent.

Better than SD3?
The minimum GPU memory required is 11GB.

Probably got censorship of some sensitive prompts such as Winnie the Pooh.

My Adobe Firefly / Generative fill is really really really bad with asian / chinese food prompts. If they can tie this into PS as a plugin would be fantastic.

 

[zheng]

Probably got censorship of some sensitive prompts such as Winnie the Pooh.

My Adobe Firefly / Generative fill is really really really bad with asian / chinese food prompts. If they can tie this into PS as a plugin would be fantastic.

i tested firefly 3 can make very nice laptop keyboard better den sdxl hahaha

 

[zheng]

Prompt:
Audrey Hepburn holding an iPhone, 1950's aesthetic, made in the 1950's, super Panavision 70 color film stock, beautiful cinematography, low contrast, balanced colors, 3 point studio lighting, high resolution, rich color palette, smooth tonality

Spoiler

too much smoothing lah losing details

 

[RyanJ]

i tested firefly 3 can make very nice laptop keyboard better den sdxl hahaha

Adobe very uptight with prompts that they think can "potentially" infringe copyrights.

Prompt:
Audrey Hepburn holding an iPhone, 1950's aesthetic, made in the 1950's, super Panavision 70 color film stock, beautiful cinematography, low contrast, balanced colors, 3 point studio lighting, high resolution, rich color palette, smooth tonality

Spoiler

The same prompt on Firefly ...

Iphone = fail; Audrey Hepburn = fail.

Cannot use names and cannot use brands.

 

[zheng]

Adobe very uptight with prompts that they think can "potentially" infringe copyrights.



The same prompt on Firefly ...

Iphone = fail; Audrey Hepburn = fail.

Cannot use names and cannot use brands.

ff3 is good if u are doing something generic to touch up your photo. can do very good portrait just no celebrity and copyright material

 

[RyanJ]

ff3 is good if u are doing something generic to touch up your photo. can do very good portrait just no celebrity and copyright material

Agree with u.

Generative fill component ( of Firefly ) is a godsend for Photoshop especially for filling up tiny gaps esp in photo stitching.

Adobe is just playing the safe game legally. They already declared their images to be trained by their ringfenced pool of Adobe licensed images so apparently no one can sue you for IP infringements.

 

[zheng]

Prompt:
Audrey Hepburn holding an iPhone, 1950's aesthetic, made in the 1950's, super Panavision 70 color film stock, beautiful cinematography, low contrast, balanced colors, 3 point studio lighting, high resolution, rich color palette, smooth tonality

steal your prompt and anyhow spam some predefined anime style on XL lightning model

breathtaking concept art Studio Ghibli anime artwork Audrey Hepburn holding an iPhone, 1950's aesthetic, smooth, NYC street, night . anime style, key visual, vibrant, studio anime, highly detailed . Anime, magical realism, environmental themes, unique characters, breathtaking landscapes, highly detailed . digital artwork, illustrative, painterly, matte painting, highly detailed . award-winning, professional, highly detailed

 

[doogyhatts]

Prompt:
1950's panavision 70 color film stock, beautiful cinematography, a giant white rabbit with pink eyes and two long ears, wears a red and white checkered coat, standing in the daisy flower field with surrounding plants and trees.

Results:
Not bad. It is able to get the basic analog film aesthetic.

Comparing with SD3.
Prompt:
A realistic giant white rabbit wearing a checkered red and white coat, standing in the flower garden, 1950s technicolor style, beautiful cinematography.

 

[zheng]

Comparing with SD3.

definitely loo better quality and nicer but not true to panvision style already

 

[doogyhatts]

definitely loo better quality and nicer but not true to panvision style already

The prompt was slightly different. Updated prompt above.

Also, I think have to use vintage filter during post-processing, because the more LoRAs I used, the 1950s technicolor style gets weaker.

 

[doogyhatts]

Haven't tried integrating ELLA into SD1.5 yet.
Now the author has reached the next phase, EMMA.

 

[zheng]

 

[doogyhatts]

DynamiCrafter can generate one foot step animation.

 

[doogyhatts]

Ok so DynamiCrafter can also generate a two foot steps animation but with a somewhat sliding motion.

Spoiler