How to Use Z-Image on a GPU with Only 4GB VRAM

[leejet]

Wiki: https://github.com/leejet/stable-diffusion.cpp/wiki/How-to-Use-Z%E2%80%90Image-on-a-GPU-with-Only-4GB-VRAM

[stduhpf]

Just a small remark: The text encoder shipped with Z-Image is not off-the-shelf Qwen3-4B-Instruct-2507. It is not base Qwen3-4B either. I believe (just a hunch) it's its own thing, they probably continued pretraining it from some variant of Qwen3 4B alongside the image gen model.

Below is a comparison between Qwen3-4B-Instruct-2507 and the "qwen_3_4B" model shipped with z-image, with f16 to avoid imatrix causing varience and in GGUF format for a fair apple-to-apple comparison.

Qwen3-4B-Instruct-2507-F16.gguf	Qwen_3_4b-FP16.gguf

[stduhpf]

Nevermind, I now think it is just base Qwen3-4B (not to be confused with Qwen3-4B-Base). Not sure why in my earlier testing I was under the impression it was a different thing.

Qwen3-4B-f16.gguf	Qwen_3_4b-FP16.gguf

[stduhpf]

I can run it at 20s/it for 512x512 resolution on my old RX 580 4GB using Vulkan on Linux. Not bad at all.

[netrunnereve]

That's suspiciously slow, on my RX 470 with Vulkan on Linux I get 7s/it at 512x512 and 35 s/it at 1024x1024 with the Q4_K model which just fits in 4GB with clip on CPU. On my 8GB card I use the Q8_0 model.

I'm using diffusion fa and vae conv direct but that shouldn't make that much of a difference.

[stduhpf]

How? I'm using Q3_K diffusion model, but that shouldn't be that much slower than Q4_k... I'm also running it with diffusion fa and vae conv direct. I was happy with my 20s/it until I read your comment.

Could be a driver issue?

Edit: with --type q4_K I get 17s/it, which is significantly faster, but still much slower than your 7s/it...
Edit 2: Reinstalled mesa drivers and Vulkan sdk, now I get 16s/it with the Q3_K model

[netrunnereve]

I just tested with Q3_K and I'm getting 7.5s/it. Something's not right and I know from when I did the Vulkan kquant optimizations that Q3 isn't that much slower than the Q4 stuff.

Are you using the default samplers? Also try to run with RADV_PERFTEST=nogttspill if your Mesa is new enough and check your clocks and vram/gtt usage with radeontop.

[stduhpf]

Also try to run with RADV_PERFTEST=nogttspill if your Mesa is new enough and check your clocks and vram/gtt usage with radeontop.

That worked! I get only ~7.3 s/it when I set this env variable.

[leejet]

On my RTX 3050 Laptop GPU, for generating 512x512 image, I can achieve 2.17s/it.

[Green-Sky]

Not very low spec, but still aging: RTX 2070, 512x512 with Q3_K_M, I get 1.19s/it.

[Green-Sky]

Worth of note is the speed of qwen on my cpu:

[DEBUG] ggml_extend.hpp:1689 - qwen3 compute buffer size: 1.86 MB(RAM)
[DEBUG] conditioner.hpp:1896 - computing condition graph completed, taking 823 ms
[INFO ] stable-diffusion.cpp:2917 - get_learned_condition completed, taking 824 ms

[engrtipusultan]

What sampler and scheduler and other settings you guys are using. I see on reddit only comfyUI. Is there a place where people are sharing their sd.cpp settings for z-image.

[engrtipusultan]

dpm2 and heun work fine too, with smoothstep and 4 or 5 steps.

Can you tell me what kind of differences you observe with these as compared to euler?

[Green-Sky]

You can indeed play with --flow-shift. The default for z-image is 3.0.

[engrtipusultan]

dpm2 and heun work fine too, with smoothstep and 4 or 5 steps.

./sd-cli --diffusion-model /home/tipu/AI/image/unet/z_image_turbo-Q8_0.gguf --vae /home/tipu/AI/image/vae/flux1vae.safetensors --llm /home/tipu/AI/image/text_encoders/Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf -p "A beautiful young woman in her early 20s lying comfortably on her bed in a softly lit modern bedroom. She has gentle, relaxed features, medium-length dark hair loosely tousled, and warm brown eyes illuminated by the soft glow of her smartphone screen as she scrolls through it. She is partially propped up against a pair of large, textured throw pillows in muted earth tones — terracotta, charcoal, and cream. The bedroom has a warm, intimate atmosphere: deep walnut furniture, matte black accents, and dim ambient lighting from a modern bedside lamp with a warm Edison bulb. The walls are painted a calming slate-gray with a few framed minimalist art prints. A narrow shelf above the bed holds small potted plants, a candle in a smoked-glass holder, and a ceramic dish with jewelry. Her bedding consists of a dark, cozy duvet in a muted olive green and a soft knitted blanket draped casually across her legs. A half-open novel rests beside her along with a pair of wireless earbuds. The wooden nightstand holds a glass of water, a charging dock with a smartwatch, and a subtle amber-glow nightlight casting gentle shadows across the room. Soft, realistic nighttime lighting dominates: the warm lamp on one side mixes with the cool, faint blue light from the smartphone, creating natural dual-tone shadows on her face. The window in the background shows blurred city lights and faint reflections in the glass. Ultra-detailed textures — the knit blanket fibers, matte ceramic surfaces, the grain of the wood furniture, faint wrinkles in the bedding — enhance the realism and atmosphere. The environment feels calm, private, mature, and lived-in." --cfg-scale 1.0 -v -H 1024 -W 1024 --steps 8 --seed -1 --vae-on-cpu --rng cpu --sampling-method dpm2 --scheduler smoothstep

--sampling-method dpm2 --scheduler smoothstep ( very slow)

Details

[wbruna]

--sampling-method dpm2 --scheduler smoothstep ( very slow)

You should reduce the step count with dpm2 / Heun: 5 (or 4) steps are equivalent to 9 (or 7) Euler.

[engrtipusultan]

dpm2 and heun work fine too, with smoothstep and 4 or 5 steps.

8 steps --sampling-method Heun --scheduler smoothstep

Details

[wbruna]

Can you tell me what kind of differences you observe with these as compared to euler?

512x512, smoothstep scheduler, same generation time (around 18s on my card):

euler 7 steps	dpm2 4 steps	heun 4 steps

[engrtipusultan]

Can you give link to one that are working for you.

crop input image from 832x1536 to 822x1536, image_path = ./output.png
resize input image from 822x1536 to 1644x3072
[WARN ] model.cpp:379  - unknown format /home/tipu/AI/image/upscale_models/4x_Nickelback_70000G.pth
[ERROR] upscaler.cpp:49   - init model loader from file failed: '/home/tipu/AI/image/upscale_models/4x_Nickelback_70000G.pth'
[INFO ] upscaler.cpp:56   - Upscaler weight type: f16
[INFO ] esrgan.hpp:170  - loading esrgan from '/home/tipu/AI/image/upscale_models/4x_Nickelback_70000G.pth'
[WARN ] model.cpp:379  - unknown format /home/tipu/AI/image/upscale_models/4x_Nickelback_70000G.pth
[ERROR] esrgan.hpp:174  - init esrgan model loader from file failed: '/home/tipu/AI/image/upscale_models/4x_Nickelback_70000G.pth'
new_upscaler_ctx failed
save result PNG image to 'upoutput.png' (success)

[wbruna]

As described on #855 , old .pth files may need to be converted. But these should work out of the box:
https://openmodeldb.info/models/4x-NickelbackFS
https://openmodeldb.info/models/4x-LSDIRplusC

[engrtipusultan]

Thank you. I will try that.

[wbruna]

I've uploaded a few converted models to huggingface: https://huggingface.co/wbruna/upscalers-sdcpp-gguf

[engrtipusultan]

Took one prompt from internet.

./sd-cli --diffusion-model /home/tipu/AI/image/unet/z_image_turbo-Q8_0.gguf --vae /home/tipu/AI/image/vae/flux1vae.safetensors --llm /home/tipu/AI/image/text_encoders/Qwen3-4B-Instruct-2507-Q8_0.gguf -p "A surreal and abstract portrait blending digital and analog textures, featuring a mysterious woman with an ethereal, otherworldly presence. She holds a smooth porcelain mask half-covering her face, its surface delicately adorned with intricate golden roses, their shimmering petals catching the light like precious metal. Her piercing, glowing golden eyes shine through the fragmented layers, exuding a sense of mystery and divine allure. Surrounding her are jagged, glitch-like fragments and torn newspaper textures, blending tactile and digital aesthetics into a disjointed yet harmonious composition. Layers of glitch art, fragmented text, and metallic streaks dominate the background, enriched by veins of radiant gold highlights that weave through the scene like threads of light. Scattered golden rose petals float through the air, their delicate form contrasting with the gritty, smudged textures and deep monochrome tones of black and grey. The interplay of gold and shadow creates a surreal, cinematic ambiance, elevating the composition with a sense of timeless beauty and fragile elegance." --cfg-scale 1.0 -v --offload-to-cpu -H 540 -W 960 --steps 7 --seed -1 --vae-on-cpu --rng cpu --scheduler smoothstep --sampling-method euler --flow-shift 2

up scaled with 4xUltrasharpV10.gguf working fine

./sd-cli -W 960 -H 544 -M upscale --upscale-model /home/tipu/AI/image/upscale_models/4xUltrasharpV10.gguf -i ./output.png -o ./upscale.png

[engrtipusultan]

I copied a prompt from internet to get painting like look.

"A Chinese female college student, gender female, age around 20. Translucent washes, wet-on-wet diffusion, soft gradients, delicate splatter effects, fluid edges"

When I do it through sd.cpp cli. I am getting variable responses. Can anyone tell me how to control it, (which parameters).

./sd-cli --diffusion-model /home/tipu/AI/image/unet/z_image_turbo-Q8_0.gguf --vae /home/tipu/AI/image/vae/UltraFlux-v1_vae.safetensors --llm /home/tipu/AI/image/text_encoders/Qwen3-4B-Instruct-2507-Q8_0.gguf --lora-model-dir /home/tipu/AI/image/loras/ --cfg-scale 1 -v -H 512 -W 512 --steps 4 --seed -1 --rng cpu --scheduler smoothstep --sampling-method heun --flow-shift 2 -o output_$(date +%Y-%m-%d_%H-%M-%S).png --vae-conv-direct --diffusion-fa -p "A Chinese female college student, gender female, age around 20. Translucent washes, wet-on-wet diffusion, soft gradients, delicate splatter effects, fluid edges "

I am getting something like this against same cli.

Most desirable Image achieved on first run (the paint splatter effect,merged properly with the original character. Character does not matter) :

Second Run, totally does not have desired effect:

Third time run, much better effect on picture as compared to second time run but not as close to the first one:

[wbruna]

When I do it through sd.cpp cli. I am getting variable responses. Can anyone tell me how to control it, (which parameters).

Just use a positive --seed parameter. For instance, --seed 366737809 will use the same seed as your first image (it's embedded on the image metadata).

[engrtipusultan]

Thank you. It almost worked. Though effect is missing on head / hair / face area but overall it is fine.

Is there any explanation anywhere how different seed values impact?

./sd-cli --diffusion-model /home/tipu/AI/image/unet/z_image_turbo-Q8_0.gguf --vae /home/tipu/AI/image/vae/UltraFlux-v1_vae.safetensors --llm /home/tipu/AI/image/text_encoders/Qwen3-4B-Instruct-2507-Q8_0.gguf --lora-model-dir /home/tipu/AI/image/loras/ --cfg-scale 1 -v -H 512 -W 512 --steps 4 --seed 366737809 --rng cpu --scheduler smoothstep --sampling-method heun --flow-shift 2 -o output_$(date +%Y-%m-%d_%H-%M-%S).png --vae-conv-direct --diffusion-fa -p "A Pakistani middle aged professor, gender male, age around 40. Translucent washes, wet-on-wet diffusion, soft gradients, delicate splatter effects, fluid edges "

[engrtipusultan]

Also is there way to keep sd.cpp running rather than each time loading the model ?

[engrtipusultan]

Another question. I am trying to add text to a picture. I see that text does not get correct spellings as mentioned in prompt and also text often is jagged.
If I remove --scheduler smoothstep then text spellings are correct and text in general is better. Is --scheduler smoothstep not to be used with writing text? or there are some other recommendations.

bash  ./sd-cli --diffusion-model /home/tipu/AI/image/unet/z_image_turbo-Q8_0.gguf --vae /home/tipu/AI/image/vae/flux1vae.safetensors --llm /home/tipu/AI/image/text_encoders/Qwen3-4B-Instruct-2507-Q8_0.gguf --lora-model-dir /home/tipu/AI/image/loras/ --cfg-scale 1 -H 1008 -W 1792 --steps 8 --seed 65464321 --sampling-method euler --flow-shift 2 -o output_$(date +%Y-%m-%d_%H-%M-%S).png --offload-to-cpu --vae-on-cpu -p "An Octane Render and C4D style city-rendered 3D digital art poster unfolds with a DreamWorks aesthetic, showcasing intricate 3D modeling and delicate details. The central focus is a majestic miniature island, precisely shaped like Islamabad's geographic outline, majestically floating above a vast expanse of pristine white clouds, dominating the frame. This aerial marvel seamlessly integrates Islamabad's iconic landmarks—the towering Faisal Mosque, the geometric Pakistan Monument, the serene Rawal Lake, and sections of the lush Margalla Hills—with architectural styles blending ancient Mughal influences into modern, clean lines, maintaining a vibrant, non-dense cityscape. City-specific birds, perhaps magpies or bulbuls, gracefully circle the island. Cinematic lighting casts long, soft shadows and sharp, sun-kissed highlights, emphasizing sunlight reflection effects across polished surfaces and verdant foliage. The island hovers over a boundless sea of clouds and distant, mist-shrouded mountains, rendered in a luminous, layered traditional South Asian artistic style with wispy, calligraphic strokes. Above this floating metropolis, the 3D text 'ISLAMABAD' appears, itself a miniature ecological installation interwoven with tiny, glowing flora and delicate, stylized vines, symbolizing cultural symbiosis. The entire vibrant scene is encircled by a minimalist, elegant information layout resembling a museum exhibit panel. It features a muted textured background with writing 'Capital of Pakistan' and 'Green Hill City', alongside fonts for auxiliary data such as 'Population: ~1.2 Million' and 'Area: 906 sq km.' In the corners, annotations in a classical atlas style, using refined typography, display '33.6844° N, 73.0479° E,' 'Founded: 1960' and 'Partly Cloudy, 25°C' all framed by extensive white space, achieving a restrained, clean, and perfectly balanced composition as if observing a treasured artifact. special attention to details. " --diffusion-fa

bash  ./sd-cli --diffusion-model /home/tipu/AI/image/unet/z_image_turbo-Q8_0.gguf --vae /home/tipu/AI/image/vae/UltraFlux-v1_vae.safetensors --llm /home/tipu/AI/image/text_encoders/Qwen3-4B-Instruct-2507-Q8_0.gguf --lora-model-dir /home/tipu/AI/image/loras/ --cfg-scale 1 -H 1280 -W 800 --steps 4 --seed -1 --rng cpu --scheduler smoothstep --sampling-method heun --flow-shift 2 -o output_$(date +%Y-%m-%d_%H-%M-%S).png --vae-conv-direct --diffusion-fa -p "An Octane Render and C4D style city-rendered 3D digital art poster unfolds with a DreamWorks aesthetic, showcasing intricate 3D modeling and delicate details. The central focus is a majestic miniature island, precisely shaped like Islamabad's geographic outline, majestically floating above a vast expanse of pristine white clouds, dominating the frame. This aerial marvel seamlessly integrates Islamabad's iconic landmarks—the towering Faisal Mosque, the geometric Pakistan Monument, the serene Rawal Lake, and sections of the lush Margalla Hills—with architectural styles blending ancient Mughal influences into modern, clean lines, maintaining a vibrant, non-dense cityscape. City-specific birds, perhaps magpies or bulbuls, gracefully circle the island. Cinematic lighting casts long, soft shadows and sharp, sun-kissed highlights, emphasizing sunlight reflection effects across polished surfaces and verdant foliage. The island hovers over a boundless sea of clouds and distant, mist-shrouded mountains, rendered in a luminous, layered traditional South Asian artistic style with wispy, calligraphic strokes. Above this floating metropolis, the 3D text "Islamabad" appears, itself a miniature ecological installation interwoven with tiny, glowing flora and delicate, stylized vines, symbolizing cultural symbiosis. The entire vibrant scene is encircled by a minimalist, elegant information layout resembling a museum exhibit panel. It features a muted textured background with classic fonts detailing 'Capital of Pakistan' and 'Green Hill City,' alongside ultra-fine minimalist fonts for auxiliary data such as 'Population: ~1.2 Million' and 'Area: 906 sq km.' In the corners, annotations in a classical atlas style, using refined serif typography, display '33.6844° N, 73.0479° E,' 'Founded: 1960,' and 'Partly Cloudy, 25°C,' all framed by extensive white space, achieving a restrained, clean, and perfectly balanced composition as if observing a treasured artifact."