HiDream-O1 Dev Instruction Edit — ComfyUI workflow

Edit one source image by a written instruction with HiDream-O1-Image Dev (bf16) 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 on (the template's shipped state): the source goes into HiDreamO1ReferenceImages as image_1 on both the prompt and the empty negative, and the output takes the source's size through GetImageSize and two ComfyMathExpression floor(a/32)*32 nodes into EmptyHiDreamO1LatentImage; sampling is the Dev chain (ModelNoiseScale 7.6, BasicScheduler normal 28 steps, SamplerLCM 1/1/2.5, SamplerCustom CFG 1). One reference is the node's instruction-edit case (its tooltip: 1 image = instruction edit, 2-10 = multi reference, which hidream-o1-personalize covers). The template's ImageScaleToTotalPixels (lanczos, 4 MP, resolution_steps 1) ships bypassed with a note to enable it for large inputs; it is always on here, so the output lands near the model's ~4 MP training size and inside the per-job pixel cap, which cannot see a size read from an image. The template loads the fp8_scaled checkpoint; this catalog loads the bf16 file on the volume. Prompt refine is not offered (see prompt_refine_note).

16Graph nodes

14Node classes

3Inputs

1Models

Tags: chromarunpod-serverlessvolume-backedimage-to-imageimage-edithidream-o1

Inputs (3)#

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

promptstringrequireddefault Transform the background into a rainy neon city street at night, with wet asphalt reflections, blurred neon signs

The edit instruction in natural prose (the default is the upstream Dev template's). 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)

source_imageimagerequired

The image being edited. The model reads it at full size as the reference (the node resizes references itself); the output takes its aspect ratio at ~4 MP (lanczos to 4 MP, each side floored to a multiple of 32, as the template computes it).

ComfyUI node graph (16)#

The executable ComfyUI prompt graph: 16 nodes across 14 distinct node classes, wired by 22 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: LoadImage"] n4["5: ImageScaleToTotalPixels"] n5["6: HiDreamO1ReferenceImages"] n6["7: GetImageSize"] n7["8: ComfyMathExpression"] n8["9: ComfyMathExpression"] n9["10: EmptyHiDreamO1LatentImage"] n10["11: ModelNoiseScale"] n11["12: BasicScheduler"] n12["13: SamplerLCM"] n13["14: SamplerCustom"] n14["15: VAEDecode"] n15["16: SaveImage"] n0 -->|clip| n1 n0 -->|clip| n2 n3 -->|image| n4 n1 -->|positive| n5 n2 -->|negative| n5 n3 -->|imagesimage_1| n5 n4 -->|image| n6 n6 -->|valuesa| n7 n6 -->|valuesa| n8 n7 -->|width| n9 n8 -->|height| n9 n0 -->|model| n10 n10 -->|model| n11 n10 -->|model| n13 n5 -->|positive| n13 n5 -->|negative| n13 n12 -->|sampler| n13 n11 -->|sigmas| n13 n9 -->|latent_image| n13 n13 -->|samples| n14 n0 -->|vae| n14 n14 -->|images| n15

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

Nodes (16)#

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

4LoadImagecore

image = {{source_image}} tmpl

outputs: IMAGEMASK

5ImageScaleToTotalPixelscore

image = ◂ node 4 · out[0]upscale_method = lanczosmegapixels = 4.0resolution_steps = 1

outputs: IMAGE

6HiDreamO1ReferenceImagescore

positive = ◂ node 2 · out[0]negative = ◂ node 3 · out[0]images.image_1 = ◂ node 4 · out[0]

outputs: CONDITIONINGCONDITIONING

7GetImageSizecore

image = ◂ node 5 · out[0]

outputs: INTINTINT

8ComfyMathExpressioncore

expression = floor(a/32)*32values.a = ◂ node 7 · out[0]

outputs: FLOATINTBOOLEAN

9ComfyMathExpressioncore

expression = floor(a/32)*32values.a = ◂ node 7 · out[1]

outputs: FLOATINTBOOLEAN

10EmptyHiDreamO1LatentImagecore

width = ◂ node 8 · out[1]height = ◂ node 9 · out[1]batch_size = 1

outputs: LATENT

11ModelNoiseScalecore

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

outputs: MODEL

12BasicSchedulercore

model = ◂ node 11 · out[0]scheduler = normalsteps = 28denoise = 1.0

outputs: SIGMAS

13SamplerLCMcore

s_noise = 1.0s_noise_end = 1.0noise_clip_std = 2.5

outputs: SAMPLER

14SamplerCustomcore

model = ◂ node 11 · out[0]add_noise = truenoise_seed = {{seed}} tmplcfg = 1.0positive = ◂ node 6 · out[0]negative = ◂ node 6 · out[1]sampler = ◂ node 13 · out[0]sigmas = ◂ node 12 · out[0]latent_image = ◂ node 10 · out[0]

outputs: LATENTLATENT

15VAEDecodecore

samples = ◂ node 14 · out[0]vae = ◂ node 1 · out[2]

outputs: IMAGE

16SaveImagecore

images = ◂ node 15 · out[0]filename_prefix = isis/hidream-o1-edit

Prompt construction#

template

{base_prompt}

Variables (1)#

base_prompt

{{prompt}} tmpl

Parameter banks (1)#

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

`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-locklatent-reuse

consistencyDimensions

identitylightingenvironment

notes

HiDream-O1-Image Dev instruction edit of one source image, output at the source aspect near 4 MP, SFW only.