HiDream-O1 Dev Text-to-Image — ComfyUI workflow

HiDream-O1-Image Dev (the distilled 8B pixel-space unified transformer, bf16) text-to-image on the RunPod image endpoint, SFW only. Translated from Comfy-Org/workflow_templates@aaac56dd templates/image_hidream_o1_dev.json with its Switch to Image Edit off: CheckpointLoaderSimple (the checkpoint carries the model, its stub text encoder and the pixel-space VAE; no external encoder or VAE), ModelNoiseScale 7.6, BasicScheduler normal at 28 steps, SamplerLCM (s_noise 1, s_noise_end 1, noise_clip_std 2.5), SamplerCustom at CFG 1 with an empty negative, EmptyHiDreamO1LatentImage at 2048x2048. The aspect options are the eleven trained resolutions the template's size note and the latent node's description list. The template loads the fp8_scaled checkpoint; this catalog loads the bf16 file on the volume. Its Enable Prompt Refine toggle (off in the template) is not offered: see prompt_refine_note.

10Graph nodes

9Node classes

4Inputs

1Models

Tags: chromarunpod-serverlessvolume-backedtext-to-imagehidream-o1

Inputs (4)#

The typed parameter surface callers bind when they request this workflow. Enum options and numeric bounds are the values the workflow document declares.

promptstringrequireddefault A dog holds a sign that says "HiDream-O1-Image release."

Positive prompt in natural prose; text in quotes is rendered as lettering (the default is the upstream HiDream-O1-Image README's Dev example). There is no negative prompt: the Dev model samples at CFG 1, where the template's empty negative is not evaluated.

seedintegerdefault -1

Random seed (-1 = random)

aspectenumdefault square_2048x2048

Output size, one of the eleven resolutions the model was trained at (the template's size note; ComfyUI v0.34.0 comfy_extras/nodes_hidream_o1.py says lower resolutions go off-distribution). All are about 4 MP and within the per-job pixel cap; 2048x2048 is the template's value.

options: square_2048x2048landscape_4x3_2304x1728portrait_3x4_1728x2304landscape_9x7_2304x1792portrait_7x9_1792x2304landscape_3x2_2496x1664portrait_2x3_1664x2496landscape_16x9_2560x1440portrait_9x16_1440x2560ultrawide_3104x1312ultratall_1312x3104

batch_sizeintegerdefault 1min 1max 1

Images per run. Pinned to 1: the executor's resolution-scaled batch cap (C.12.19) allows a second latent only below ~0.5 MP, and every trained resolution is ~4 MP.

ComfyUI node graph (10)#

The executable ComfyUI prompt graph: 10 nodes across 9 distinct node classes, wired by 13 data dependencies. Nodes tinted green come from a custom node pack this workflow declares; the rest are ComfyUI core / baked-community classes.

flowchart TD n0["1: CheckpointLoaderSimple"] n1["2: CLIPTextEncode"] n2["3: CLIPTextEncode"] n3["4: ModelNoiseScale"] n4["5: BasicScheduler"] n5["6: SamplerLCM"] n6["7: EmptyHiDreamO1LatentImage"] n7["8: SamplerCustom"] n8["9: VAEDecode"] n9["10: SaveImage"] n0 -->|clip| n1 n0 -->|clip| n2 n0 -->|model| n3 n3 -->|model| n4 n3 -->|model| n7 n1 -->|positive| n7 n2 -->|negative| n7 n5 -->|sampler| n7 n4 -->|sigmas| n7 n6 -->|latent_image| n7 n7 -->|samples| n8 n0 -->|vae| n8 n8 -->|images| n9

Data-flow DAG — scroll to zoom, drag to pan.

Nodes (10)#

1CheckpointLoaderSimplecore

ckpt_name = hidream_o1_image_dev_bf16.safetensors

outputs: MODELCLIPVAE

2CLIPTextEncodecore

text = {{constructed_prompt}} tmplclip = ◂ node 1 · out[1]

outputs: CONDITIONING

3CLIPTextEncodecore

text = clip = ◂ node 1 · out[1]

outputs: CONDITIONING

4ModelNoiseScalecore

model = ◂ node 1 · out[0]noise_scale = 7.6

outputs: MODEL

5BasicSchedulercore

model = ◂ node 4 · out[0]scheduler = normalsteps = 28denoise = 1.0

outputs: SIGMAS

6SamplerLCMcore

s_noise = 1.0s_noise_end = 1.0noise_clip_std = 2.5

outputs: SAMPLER

7EmptyHiDreamO1LatentImagecore

width = {{aspect_map[aspect].width}} tmplheight = {{aspect_map[aspect].height}} tmplbatch_size = {{batch_size}} tmpl

outputs: LATENT

8SamplerCustomcore

model = ◂ node 4 · out[0]add_noise = truenoise_seed = {{seed}} tmplcfg = 1.0positive = ◂ node 2 · out[0]negative = ◂ node 3 · out[0]sampler = ◂ node 6 · out[0]sigmas = ◂ node 5 · out[0]latent_image = ◂ node 7 · out[0]

outputs: LATENTLATENT

9VAEDecodecore

samples = ◂ node 8 · out[0]vae = ◂ node 1 · out[2]

outputs: IMAGE

10SaveImagecore

images = ◂ node 9 · out[0]filename_prefix = isis/hidream-o1-txt2img

Prompt construction#

template

{base_prompt}

Variables (1)#

base_prompt

{{prompt}} tmpl

Parameter banks (2)#

The prompt / configuration lookup tables this workflow keys into from its inputs — the vocabulary that turns a style / palette / preset selection into graph parameters.

`aspect_map` (11)#

square_2048x2048

{"width": 2048, "height": 2048}

landscape_4x3_2304x1728

{"width": 2304, "height": 1728}

portrait_3x4_1728x2304

{"width": 1728, "height": 2304}

landscape_9x7_2304x1792

{"width": 2304, "height": 1792}

portrait_7x9_1792x2304

{"width": 1792, "height": 2304}

landscape_3x2_2496x1664

{"width": 2496, "height": 1664}

portrait_2x3_1664x2496

{"width": 1664, "height": 2496}

landscape_16x9_2560x1440

{"width": 2560, "height": 1440}

portrait_9x16_1440x2560

{"width": 1440, "height": 2560}

ultrawide_3104x1312

{"width": 3104, "height": 1312}

ultratall_1312x3104

{"width": 1312, "height": 3104}

`requires_families` (1)#

hidream-o1

Models & dependencies#

Models required (1)#

hidream_o1_image_dev_bf16.safetensors

Output contract#

What a successful run of this workflow returns.

type

image

format

png

Taxonomy & routing#

How the control plane classifies this workflow — from the committed workflow-taxonomy-registry.json. It drives the consistency / control surface the agentic director can exercise over the workflow.

assetFamily

portrait-hero-image-bundle

outputPackageProfile

image-single-profile

controlModalities

model-locksampler-scheduler-lockseed-lockprompt-template-lock

consistencyDimensions

identitylightinglens

notes

HiDream-O1-Image Dev text-to-image at its trained ~4 MP resolutions on RunPod serverless, SFW only.