Instructions to use vidfom/Ltx-3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- llama-cpp-python
How to use vidfom/Ltx-3 with llama-cpp-python:
# !pip install llama-cpp-python from llama_cpp import Llama llm = Llama.from_pretrained( repo_id="vidfom/Ltx-3", filename="ComfyUI/models/text_encoders/gemma-3-12b-it-qat-UD-Q4_K_XL.gguf", )
llm.create_chat_completion( messages = "No input example has been defined for this model task." )
- Notebooks
- Google Colab
- Kaggle
- Local Apps
- llama.cpp
How to use vidfom/Ltx-3 with llama.cpp:
Install from brew
brew install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama-server -hf vidfom/Ltx-3:UD-Q4_K_XL # Run inference directly in the terminal: llama-cli -hf vidfom/Ltx-3:UD-Q4_K_XL
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama-server -hf vidfom/Ltx-3:UD-Q4_K_XL # Run inference directly in the terminal: llama-cli -hf vidfom/Ltx-3:UD-Q4_K_XL
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf vidfom/Ltx-3:UD-Q4_K_XL # Run inference directly in the terminal: ./llama-cli -hf vidfom/Ltx-3:UD-Q4_K_XL
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf vidfom/Ltx-3:UD-Q4_K_XL # Run inference directly in the terminal: ./build/bin/llama-cli -hf vidfom/Ltx-3:UD-Q4_K_XL
Use Docker
docker model run hf.co/vidfom/Ltx-3:UD-Q4_K_XL
- LM Studio
- Jan
- Ollama
How to use vidfom/Ltx-3 with Ollama:
ollama run hf.co/vidfom/Ltx-3:UD-Q4_K_XL
- Unsloth Studio new
How to use vidfom/Ltx-3 with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for vidfom/Ltx-3 to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for vidfom/Ltx-3 to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for vidfom/Ltx-3 to start chatting
- Docker Model Runner
How to use vidfom/Ltx-3 with Docker Model Runner:
docker model run hf.co/vidfom/Ltx-3:UD-Q4_K_XL
- Lemonade
How to use vidfom/Ltx-3 with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull vidfom/Ltx-3:UD-Q4_K_XL
Run and chat with the model
lemonade run user.Ltx-3-UD-Q4_K_XL
List all available models
lemonade list
File size: 13,227 Bytes
e00eceb | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 | import os
import uuid
from pathlib import Path
import pytest
import requests
from helpers import get_asset_filename, trigger_sync_seed_assets
@pytest.mark.parametrize("root", ["input", "output"])
def test_seed_asset_removed_when_file_is_deleted(
root: str,
http: requests.Session,
api_base: str,
comfy_tmp_base_dir: Path,
):
"""Asset without hash (seed) whose file disappears:
after triggering sync_seed_assets, Asset + AssetInfo disappear.
"""
# Create a file directly under input/unit-tests/<case> so tags include "unit-tests"
case_dir = comfy_tmp_base_dir / root / "unit-tests" / "syncseed"
case_dir.mkdir(parents=True, exist_ok=True)
name = f"seed_{uuid.uuid4().hex[:8]}.bin"
fp = case_dir / name
fp.write_bytes(b"Z" * 2048)
# Trigger a seed sync so DB sees this path (seed asset => hash is NULL)
trigger_sync_seed_assets(http, api_base)
# Verify it is visible via API and carries no hash (seed)
r1 = http.get(
api_base + "/api/assets",
params={"include_tags": "unit-tests,syncseed", "name_contains": name},
timeout=120,
)
body1 = r1.json()
assert r1.status_code == 200
# there should be exactly one with that name
matches = [a for a in body1.get("assets", []) if a.get("name") == name]
assert matches
assert matches[0].get("asset_hash") is None
asset_info_id = matches[0]["id"]
# Remove the underlying file and sync again
if fp.exists():
fp.unlink()
trigger_sync_seed_assets(http, api_base)
# It should disappear (AssetInfo and seed Asset gone)
r2 = http.get(
api_base + "/api/assets",
params={"include_tags": "unit-tests,syncseed", "name_contains": name},
timeout=120,
)
body2 = r2.json()
assert r2.status_code == 200
matches2 = [a for a in body2.get("assets", []) if a.get("name") == name]
assert not matches2, f"Seed asset {asset_info_id} should be gone after sync"
@pytest.mark.skip(reason="Requires computing hashes of files in directories to verify and clear missing tags")
def test_hashed_asset_missing_tag_added_then_removed_after_scan(
http: requests.Session,
api_base: str,
comfy_tmp_base_dir: Path,
asset_factory,
make_asset_bytes,
):
"""Hashed asset with a single cache_state:
1. delete its file -> sync adds 'missing'
2. restore file -> sync removes 'missing'
"""
name = "missing_tag_test.png"
tags = ["input", "unit-tests", "msync2"]
data = make_asset_bytes(name, 4096)
a = asset_factory(name, tags, {}, data)
# Compute its on-disk path and remove it
dest = comfy_tmp_base_dir / "input" / "unit-tests" / "msync2" / get_asset_filename(a["asset_hash"], ".png")
assert dest.exists(), f"Expected asset file at {dest}"
dest.unlink()
# Fast sync should add 'missing' to the AssetInfo
trigger_sync_seed_assets(http, api_base)
g1 = http.get(f"{api_base}/api/assets/{a['id']}", timeout=120)
d1 = g1.json()
assert g1.status_code == 200, d1
assert "missing" in set(d1.get("tags", [])), "Expected 'missing' tag after deletion"
# Restore the file with the exact same content and sync again
dest.parent.mkdir(parents=True, exist_ok=True)
dest.write_bytes(data)
trigger_sync_seed_assets(http, api_base)
g2 = http.get(f"{api_base}/api/assets/{a['id']}", timeout=120)
d2 = g2.json()
assert g2.status_code == 200, d2
assert "missing" not in set(d2.get("tags", [])), "Missing tag should be cleared after verify"
def test_hashed_asset_two_asset_infos_both_get_missing(
http: requests.Session,
api_base: str,
comfy_tmp_base_dir: Path,
asset_factory,
):
"""Hashed asset with a single cache_state, but two AssetInfo rows:
deleting the single file then syncing should add 'missing' to both infos.
"""
# Upload one hashed asset
name = "two_infos_one_path.png"
base_tags = ["input", "unit-tests", "multiinfo"]
created = asset_factory(name, base_tags, {}, b"A" * 2048)
# Create second AssetInfo for the same Asset via from-hash
payload = {
"hash": created["asset_hash"],
"name": "two_infos_one_path_copy.png",
"tags": base_tags, # keep it in our unit-tests scope for cleanup
"user_metadata": {"k": "v"},
}
r2 = http.post(api_base + "/api/assets/from-hash", json=payload, timeout=120)
b2 = r2.json()
assert r2.status_code == 201, b2
second_id = b2["id"]
# Remove the single underlying file
p = comfy_tmp_base_dir / "input" / "unit-tests" / "multiinfo" / get_asset_filename(b2["asset_hash"], ".png")
assert p.exists()
p.unlink()
r0 = http.get(api_base + "/api/tags", params={"limit": "1000", "include_zero": "false"}, timeout=120)
tags0 = r0.json()
assert r0.status_code == 200, tags0
byname0 = {t["name"]: t for t in tags0.get("tags", [])}
old_missing = int(byname0.get("missing", {}).get("count", 0))
# Sync -> both AssetInfos for this asset must receive 'missing'
trigger_sync_seed_assets(http, api_base)
ga = http.get(f"{api_base}/api/assets/{created['id']}", timeout=120)
da = ga.json()
assert ga.status_code == 200, da
assert "missing" in set(da.get("tags", []))
gb = http.get(f"{api_base}/api/assets/{second_id}", timeout=120)
db = gb.json()
assert gb.status_code == 200, db
assert "missing" in set(db.get("tags", []))
# Tag usage for 'missing' increased by exactly 2 (two AssetInfos)
r1 = http.get(api_base + "/api/tags", params={"limit": "1000", "include_zero": "false"}, timeout=120)
tags1 = r1.json()
assert r1.status_code == 200, tags1
byname1 = {t["name"]: t for t in tags1.get("tags", [])}
new_missing = int(byname1.get("missing", {}).get("count", 0))
assert new_missing == old_missing + 2
@pytest.mark.skip(reason="Requires computing hashes of files in directories to deduplicate into multiple cache states")
def test_hashed_asset_two_cache_states_partial_delete_then_full_delete(
http: requests.Session,
api_base: str,
comfy_tmp_base_dir: Path,
asset_factory,
make_asset_bytes,
run_scan_and_wait,
):
"""Hashed asset with two cache_state rows:
1. delete one file -> sync should NOT add 'missing'
2. delete second file -> sync should add 'missing'
"""
name = "two_cache_states_partial_delete.png"
tags = ["input", "unit-tests", "dual"]
data = make_asset_bytes(name, 3072)
created = asset_factory(name, tags, {}, data)
path1 = comfy_tmp_base_dir / "input" / "unit-tests" / "dual" / get_asset_filename(created["asset_hash"], ".png")
assert path1.exists()
# Create a second on-disk copy under the same root but different subfolder
path2 = comfy_tmp_base_dir / "input" / "unit-tests" / "dual_copy" / name
path2.parent.mkdir(parents=True, exist_ok=True)
path2.write_bytes(data)
# Fast seed so the second path appears (as a seed initially)
trigger_sync_seed_assets(http, api_base)
# Deduplication of AssetInfo-s will not happen as first AssetInfo has owner='default' and second has empty owner.
run_scan_and_wait("input")
# Remove only one file and sync -> asset should still be healthy (no 'missing')
path1.unlink()
trigger_sync_seed_assets(http, api_base)
g1 = http.get(f"{api_base}/api/assets/{created['id']}", timeout=120)
d1 = g1.json()
assert g1.status_code == 200, d1
assert "missing" not in set(d1.get("tags", [])), "Should not be missing while one valid path remains"
# Baseline 'missing' usage count just before last file removal
r0 = http.get(api_base + "/api/tags", params={"limit": "1000", "include_zero": "false"}, timeout=120)
tags0 = r0.json()
assert r0.status_code == 200, tags0
old_missing = int({t["name"]: t for t in tags0.get("tags", [])}.get("missing", {}).get("count", 0))
# Remove the second (last) file and sync -> now we expect 'missing' on this AssetInfo
path2.unlink()
trigger_sync_seed_assets(http, api_base)
g2 = http.get(f"{api_base}/api/assets/{created['id']}", timeout=120)
d2 = g2.json()
assert g2.status_code == 200, d2
assert "missing" in set(d2.get("tags", [])), "Missing must be set once no valid paths remain"
# Tag usage for 'missing' increased by exactly 2 (two AssetInfo for one Asset)
r1 = http.get(api_base + "/api/tags", params={"limit": "1000", "include_zero": "false"}, timeout=120)
tags1 = r1.json()
assert r1.status_code == 200, tags1
new_missing = int({t["name"]: t for t in tags1.get("tags", [])}.get("missing", {}).get("count", 0))
assert new_missing == old_missing + 2
@pytest.mark.parametrize("root", ["input", "output"])
def test_missing_tag_clears_on_fastpass_when_mtime_and_size_match(
root: str,
http: requests.Session,
api_base: str,
comfy_tmp_base_dir: Path,
asset_factory,
make_asset_bytes,
):
"""
Fast pass alone clears 'missing' when size and mtime match exactly:
1) upload (hashed), record original mtime_ns
2) delete -> fast pass adds 'missing'
3) restore same bytes and set mtime back to the original value
4) run fast pass again -> 'missing' is removed (no slow scan)
"""
scope = f"fastclear-{uuid.uuid4().hex[:6]}"
name = "fastpass_clear.bin"
data = make_asset_bytes(name, 3072)
a = asset_factory(name, [root, "unit-tests", scope], {}, data)
aid = a["id"]
base = comfy_tmp_base_dir / root / "unit-tests" / scope
p = base / get_asset_filename(a["asset_hash"], ".bin")
st0 = p.stat()
orig_mtime_ns = getattr(st0, "st_mtime_ns", int(st0.st_mtime * 1_000_000_000))
# Delete -> fast pass adds 'missing'
p.unlink()
trigger_sync_seed_assets(http, api_base)
g1 = http.get(f"{api_base}/api/assets/{aid}", timeout=120)
d1 = g1.json()
assert g1.status_code == 200, d1
assert "missing" in set(d1.get("tags", []))
# Restore same bytes and revert mtime to the original value
p.parent.mkdir(parents=True, exist_ok=True)
p.write_bytes(data)
# set both atime and mtime in ns to ensure exact match
os.utime(p, ns=(orig_mtime_ns, orig_mtime_ns))
# Fast pass should clear 'missing' without a scan
trigger_sync_seed_assets(http, api_base)
g2 = http.get(f"{api_base}/api/assets/{aid}", timeout=120)
d2 = g2.json()
assert g2.status_code == 200, d2
assert "missing" not in set(d2.get("tags", [])), "Fast pass should clear 'missing' when size+mtime match"
@pytest.mark.skip(reason="Requires computing hashes of files in directories to deduplicate into multiple cache states")
@pytest.mark.parametrize("root", ["input", "output"])
def test_fastpass_removes_stale_state_row_no_missing(
root: str,
http: requests.Session,
api_base: str,
comfy_tmp_base_dir: Path,
asset_factory,
make_asset_bytes,
run_scan_and_wait,
):
"""
Hashed asset with two states:
- delete one file
- run fast pass only
Expect:
- asset stays healthy (no 'missing')
- stale AssetCacheState row for the deleted path is removed.
We verify this behaviorally by recreating the deleted path and running fast pass again:
a new *seed* AssetInfo is created, which proves the old state row was not reused.
"""
scope = f"stale-{uuid.uuid4().hex[:6]}"
name = "two_states.bin"
data = make_asset_bytes(name, 2048)
# Upload hashed asset at path1
a = asset_factory(name, [root, "unit-tests", scope], {}, data)
base = comfy_tmp_base_dir / root / "unit-tests" / scope
a1_filename = get_asset_filename(a["asset_hash"], ".bin")
p1 = base / a1_filename
assert p1.exists()
aid = a["id"]
h = a["asset_hash"]
# Create second state path2, seed+scan to dedupe into the same Asset
p2 = base / "copy" / name
p2.parent.mkdir(parents=True, exist_ok=True)
p2.write_bytes(data)
trigger_sync_seed_assets(http, api_base)
run_scan_and_wait(root)
# Delete path1 and run fast pass -> no 'missing' and stale state row should be removed
p1.unlink()
trigger_sync_seed_assets(http, api_base)
g1 = http.get(f"{api_base}/api/assets/{aid}", timeout=120)
d1 = g1.json()
assert g1.status_code == 200, d1
assert "missing" not in set(d1.get("tags", []))
# Recreate path1 and run fast pass again.
# If the stale state row was removed, a NEW seed AssetInfo will appear for this path.
p1.write_bytes(data)
trigger_sync_seed_assets(http, api_base)
rl = http.get(
api_base + "/api/assets",
params={"include_tags": f"unit-tests,{scope}"},
timeout=120,
)
bl = rl.json()
assert rl.status_code == 200, bl
items = bl.get("assets", [])
# one hashed AssetInfo (asset_hash == h) + one seed AssetInfo (asset_hash == null)
hashes = [it.get("asset_hash") for it in items if it.get("name") in (name, a1_filename)]
assert h in hashes
assert any(x is None for x in hashes), "Expected a new seed AssetInfo for the recreated path"
# Asset identity still healthy
rh = http.head(f"{api_base}/api/assets/hash/{h}", timeout=120)
assert rh.status_code == 200
|