Paper · 2501.18623 · ICLR · 2025
VLMaterial: Procedural Material Generation with Large Vision-Language Models
Armando Solar-Lezama, Wojciech Matusik, Rundi Wu, Changxi Zheng, Beichen Li, Liang Shi, Bernd Bickel
arXiv · PDF · Open in the Atlas
Code that ran
3 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 1 failure of our instrument; 1 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.
- 2 ran with no instrument failure
0 honoured the contract we drafted, 0 violated it, 2 have no contract verdict
- 1 ran where Syntology's instrument failed
failures of Syntology's instrument, not of the code: 1 where our contract draft was wrong, 0 where the contract check could not run
- 1 have no recorded run
- 2 of those that ran constructed an object rather than computing a result
an executed sample whose run returned an instance of its own class (fixture_out_type equals the entry name): the run built an object and did not compute a result
- 0 are pointer only: the body is not served under this paper's link
"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2501.18623.json · badge: /api/ran/2501.18623.svg
Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.
| Repository | Role | Ran |
|---|
| mit-gfx/VLMaterial |
community Papers with Code archive |
3 ran · 1 no recorded run |
| Function | Status | Where it lives |
|---|
| TaskInfo |
Ran ran on a synthesized input · constructed an object; no result was computed |
mit-gfx/VLMaterial/data_scripts/gen_programs_llm_async.py code served (permissive licence) · get_code("1bc38c9879b31f4c") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, TaskInfo: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
import torch
from dataclasses import dataclass
torch.manual_seed(0)
@dataclass
class TaskInfo:
"""Information about a generation task.
"""
case_id: int
sample_id: int
return_id: int
files: tuple[str, str]
output_folder: str
ARGS = {
"case_id": 42,
"sample_id": 7,
"return_id": 13,
"files": ("input_file.txt", "output_file.txt"),
"output_folder": "/tmp/results"
}
- Setup block drafted by
- a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
- Imports
- torch — import names as written, not package names to install; a name the repository itself provides is not a package
- Returned
- type TaskInfo — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a TaskInfo output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
- Environment
- image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
|
| TaskManager |
Ran ran on a synthesized input · constructed an object; no result was computed |
mit-gfx/VLMaterial/data_scripts/gen_programs_llm_async.py code served (permissive licence) · get_code("04023c47246259e9") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, TaskManager: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
from argparse import ArgumentParser, Namespace
import tempfile
import os
# Create a temporary directory for output
temp_dir = tempfile.mkdtemp()
# Create a minimal dataset (list of file paths)
dataset = [
"/fake/path/file1.txt",
"/fake/path/file2.txt",
"/fake/path/file3.txt",
"/fake/path/file4.txt",
]
# Create args Namespace with all required attributes
args = Namespace(
dataset=dataset,
num_cases=3,
num_samples=5,
max_samples=10,
seed=42,
output_folder=temp_dir,
resume=False,
)
# Define ARGS dictionary
ARGS = {
"dataset": dataset,
"args": args,
}
- Setup block drafted by
- a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
- Imports
- numpy, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
- Returned
- type TaskManager — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a TaskManager output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
- Environment
- image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
|
| make_prompt |
Ran ran; our contract draft was wrong, not the code |
mit-gfx/VLMaterial/data_scripts/gen_programs_llm_async.py code served (permissive licence) · get_code("3884ccd0ec99644e") How it ranRan in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.
import torch
import numpy as np
SYSTEM_PROMPT_V2 = "You are familiar with creating procedural materials using Blender's Python API. You will be given a Python code block delimited by triple backticks, which contains two functions that define the node trees of two procedural materials in Blender. The functions are named `shader_material_1` and `shader_material_2`, respectively. Your task is to write a Python function `shader_material` that creates a new material based on the provided functions. Write your code following the guidelines below.\n\nCode template:\n```python\nimport bpy\n\ndef shader_material(material: bpy.types.Material):\n material.use_nodes = True\n nodes = material.node_tree.nodes\n links = material.node_tree.links\n\n # Create nodes\n # YOUR CODE HERE\n\n # Create links to connect nodes\n # YOUR CODE HERE\n\n # Set parameters for each node\n # YOUR CODE HERE\n```\n\nRules:\n1. Create no more than 30 nodes. Only use node types and parameter fields referenced in the provided functions. Change parameter values as needed.\n2. Make sure your code can be correctly executed in Blender 3.3. Refer to the Blender Python API documentation for valid node types and parameters.\n3. Try to generate materials with complex and semantically meaningful appearances. Avoid generating materials that lack structure or look too similar to the provided examples.\n4. Follow the format and coding style of the example functions. Do not add new code blocks or comments.\n5. Simply reply with code. Exclude any additional text or explanations.\n"
func1_code = """
def shader_material_1(material):
material.use_nodes = True
nodes = material.node_tree.nodes
links = material.node_tree.links
nodes.clear()
node_principled = nodes.new(type='ShaderNodeBsdfPrincipled')
node_output = nodes.new(type='ShaderNodeOutputMaterial')
node_principled.inputs['Base Color'].default_value = (0.8, 0.2, 0.1, 1.0)
node_principled.inputs['Roughness'].default_value = 0.5
links.new(node_principled.outputs['BSDF'], node_output.inputs['Surface'])
"""
func2_code = """
def shader_material_2(material):
material.use_nodes = True
nodes = material.node_tree.nodes
links = material.node_tree.links
nodes.clear()
node_noise = nodes.new(type='ShaderNodeTexNoise')
node_color_ramp = nodes.new(type='ShaderNodeValRamp')
node_principled = nodes.new(type='ShaderNodeBsdfPrincipled')
node_output = nodes.new(type='ShaderNodeOutputMaterial')
node_noise.inputs['Scale'].default_value = 15.0
node_color_ramp.color_ramp.elements[0].color = (0.0, 0.0, 0.0, 1.0)
node_color_ramp.color_ramp.elements[1].color = (1.0, 1.0, 1.0, 1.0)
node_principled.inputs['Base Color'].default_value = (0.5, 0.5, 0.5, 1.0)
links.new(node_noise.outputs['Fac'], node_color_ramp.inputs['Fac'])
links.new(node_color_ramp.outputs['Color'], node_principled.inputs['Base Color'])
links.new(node_principled.outputs['BSDF'], node_output.inputs['Surface'])
"""
ARGS = {
'func1': func1_code,
'func2': func2_code
}
- Setup block drafted by
- a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
- Imports
- numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
- Returned
- type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
- dtype family
- int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
- Environment
- image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
|
| gen_worker |
Not yet run unverified (harvested, no recorded run) |
mit-gfx/VLMaterial/data_scripts/gen_programs_llm_async.py code served (permissive licence) · get_code("37778e1b304336c0") How it ran |
Also present in other papers' code
Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 4 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).
Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.
Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.
Repositories linked to this paper
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Abstract
Procedural materials, represented as functional node graphs, are ubiquitous in computer graphics for photorealistic material appearance design. They allow users to perform intuitive and precise editing to achieve desired visual appearances. However, creating a procedural material given an input image requires professional knowledge and significant effort. In this work, we leverage the ability to convert procedural materials into standard Python programs and fine-tune a large pre-trained vision-language model (VLM) to generate such programs from input images. To enable effective fine-tuning, we also contribute an open-source procedural material dataset and propose to perform program-level augmentation by prompting another pre-trained large language model (LLM). Through extensive evaluation, we show that our method outperforms previous methods on both synthetic and real-world examples.
For agents
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2501.18623")
get_code_for_paper("2501.18623")
have("2501.18623")
The run record, dated, one paper per request, free:
curl https://syntology.ai/api/ran/2501.18623.json
A badge for a README (the split and the date, never a ratio):
[](https://syntology.ai/paper/2501.18623)
Connect an agent — have() is free.