z-Image Base model

z-Image Base – K-Sampler Steps: 25, Seed: 603859269672671, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.

z-Image Turboには衝撃を受けたのですが、なんか同じ顔の女の子しか出てきませんよ?という気がしてました。一月前くらいに、蒸留モデルじゃないz-Image Baseというのが出たのでためしてみました。 

z-Image Turboで使ったComfy-UIのWorkflowをそのまま使って、BaseモデルだとK-Samplerというやつのステップ数を増やしたほうがいいらしいので、増やしてみて(9になってたのを25にしました)。出てきたのが上の画像。お顔はいいけど手とかコーヒーカップとかが破綻してます(笑)。Turboで生成すると、

z-Image Base – K-Sampler Steps: 25, Seed: 603859269672671, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.

手もコーヒーカップもちゃんとしてますが、なんか見覚えのある顔だなー、という印象です。

Seedを1ずつ増やして画像生成すると違いが歴然。まずはz-Image Basモデル。

z-Image Base – K-Sampler Steps: 25, Seed: 603859269672674, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.
z-Image Base – K-Sampler Steps: 25, Seed: 603859269672675, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.
z-Image Base – K-Sampler Steps: 25, Seed: 603859269672676, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.
z-Image Base – K-Sampler Steps: 25, Seed: 603859269672677, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.
z-Image Base – K-Sampler Steps: 25, Seed: 603859269672678, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.

まぁ背景とか手とかカップはかなりめちゃくちゃですが、とりあえずいろんなタイプの顔の画像が生成されて、楽しい。これが、z-Image Turboになると、

z-Image Turbo – K-Sampler Steps: 25, Seed: 603859269672674, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.
z-Image Base – K-Sampler Steps: 25, Seed: 603859269672675, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.
z-Image Base – K-Sampler Steps: 25, Seed: 603859269672676, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.
z-Image Base – K-Sampler Steps: 25, Seed: 603859269672677, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.
z-Image Base – K-Sampler Steps: 25, Seed: 603859269672678, Prompt: Full body shot, 29 year old Japanese woman, IT professional, short choppy light brown hair, plain beige oversized T-shirt, thin arms, sitting at coffee shop counter, holding cup, looking at viewer, direct eye contact. Detailed face, small nose, tiny pink lips, natural skin texture, visible pores, faint crow’s feet, unretouched, raw photo. Dim moody lighting, sister vibe, high quality, photorealistic, 8k.

手やカップも破綻してないし、背景もかっこよくボケていて、Baseモデルが生成した画像と比べると、間違いなくこっちのTurboモデルの生成画像の方がクオリティが高い。でも、同じ顔ばっかりですな。このモデルさんには悪いけど(AIにあやまる必要はないか、笑)。

Leave a comment