# When is Renest the wrong tool?

> When the problem isn't moving a setup that already works onto another Linux GPU machine. Housekeeping on one PC, a Windows-to-Windows move, a first install that never worked, a production deployment, or a framework other than ComfyUI, kohya_ss or LLaMA-Factory all have better answers, and this page gives them.

Canonical page: https://renest.ai/answers/when-you-dont-need-renest

<h2 id="what-it-is-for">What Renest is for, in one paragraph</h2>

<p>Renest is for one job: you have ComfyUI, kohya_ss or LLaMA-Factory working on a Linux
machine with an NVIDIA GPU, usually a rented one, and you want that working setup back on
another machine. It keeps the files and a reinstallable record of the dependencies, checks
every file when it comes back, and tests whether the app works on the new machine. If your
problem is anything else, one of the answers below is simpler.</p>

<h2 id="another-drive">I want to move my models to another drive on the same PC</h2>

<p>Point ComfyUI at the new folder; you don't need to move the install. Two ways:</p>

<ul>
  <li><strong><code>extra_model_paths.yaml</code>.</strong> In the ComfyUI folder, copy
  <code>extra_model_paths.yaml.example</code> to <code>extra_model_paths.yaml</code> and add
  a section whose <code>base_path</code> is the new location, with one line per model type,
  for example <code>checkpoints: models/checkpoints/</code>. Restart ComfyUI.</li>
  <li><strong>A directory junction on Windows.</strong> Move the <code>models</code> folder to
  the new drive, then create a junction where it used to be, so ComfyUI sees the same path.
  From a Command Prompt:</li>
</ul>

<pre>$ mklink /J "C:\ComfyUI\models" "D:\ComfyUI-models"</pre>

<p>The first path is where ComfyUI expects the folder, and must not exist any more; the
second is where the files now live. On Linux, a symbolic link
(<code>ln -s</code>) does the same.</p>

<h2 id="updates">I want my models and workflows to survive ComfyUI updates</h2>

<p>Keep them outside the install, so an update only replaces the program:</p>

<ul>
  <li>Models: point <code>extra_model_paths.yaml</code> at a folder outside the install, as
  above.</li>
  <li>Workflows and settings: recent ComfyUI versions have launch options that move the
  user, output and input folders outside the install; <code>python main.py -h</code> lists
  them.</li>
  <li>Custom nodes have to stay inside each install, in its own <code>custom_nodes</code>
  folder, with their packages in that install's Python.</li>
  <li>Before a big update, keep a copy of the whole program folder. Going back is then a
  rename.</li>
</ul>

<h2 id="shared-models">I run several ComfyUI installs and want them to share models</h2>

<p>Models can be shared; custom nodes can't. Point every install's
<code>extra_model_paths.yaml</code> at one model folder. Each install still loads custom
nodes from its own <code>custom_nodes</code> folder and needs their packages in its own
Python. Pointing a second install at the first one's node folder doesn't install those
packages there.</p>

<h2 id="reinstall">I clicked reinstall. Are my models and custom nodes gone?</h2>

<p>Move <code>custom_nodes</code>, <code>models</code> and <code>user</code> out of the
install folder before the reinstall touches them, and move them back afterwards. Models
usually work at once. Custom nodes often need their packages installed again, because the
reinstall replaced the Python they lived in; ComfyUI-Manager or the node's
<code>requirements.txt</code> does that. If the reinstall has already overwritten or deleted
the files, there is no undo. Restore them from a backup if you have one.</p>

<h2 id="portable-pip">I ran pip install and ComfyUI Portable still says the package is missing</h2>

<p>The package went into a different Python. ComfyUI Portable runs its own embedded Python,
and a bare <code>pip</code> or <code>python -m pip</code> installs into whichever Python is
first on your path. From the <code>ComfyUI_windows_portable</code> folder, use the embedded
one:</p>

<pre>$ python_embeded\python.exe -s -m pip install &lt;package&gt;</pre>

<p>If that Python has no pip at all, ComfyUI-Manager can't install anything either. Install
pip into that same <code>python.exe</code>, then restart.</p>

<h2 id="windows">I'm moving ComfyUI from one Windows PC to another</h2>

<p>Copy the whole Portable folder. The embedded Python, the custom nodes and their packages
are all inside it, so it usually runs on the new PC as long as the GPU is from a family the
same PyTorch build supports. Copy the model folder too, if it lives elsewhere, and update
<code>extra_model_paths.yaml</code> if its location changed. Renest restores onto Linux
machines with an NVIDIA GPU, and the desktop app isn't built yet, so it doesn't help with this
move.</p>

<h2 id="first-install">ComfyUI, kohya_ss or LLaMA-Factory has never worked for me</h2>

<p>Renest can't help with a first install. It only carries a setup from a run that already
worked: it doesn't resolve dependency conflicts, choose versions, or recommend models,
templates or training settings. Get it working once with the project's own install guide or
a ready-made template on your GPU provider. After that, Renest can keep the working version
for the next machine.</p>

<h2 id="serverless">I want a serverless endpoint or a production deployment</h2>

<p>Build a Docker image with your dependencies installed, and keep the models on a volume or
download them at start-up. That is what serverless GPU platforms expect. Renest isn't a
deployment platform. It restores a setup onto a GPU machine you rent, for you to work
on.</p>

<h2 id="frameworks">I use A1111, Forge, diffusion-pipe, musubi-tuner, ai-toolkit or Axolotl</h2>

<p>Renest doesn't support them. It is built for three named frameworks: ComfyUI for image
generation, and kohya_ss and LLaMA-Factory for fine-tuning. The checks described in these
docs, such as which custom nodes loaded or whether training wrote a file, are written for
those three, and we don't test anything else.</p>

<h2 id="other">Other things Renest doesn't do</h2>

<ul>
  <li><strong>Make the same seed give the same image on another GPU.</strong> Files and
  versions come back as they were, and different hardware can still produce different
  pixels. Why: <a href="/blog/cpu-vendor-decides-identical-images">Same GPU, two
  images</a>.</li>
  <li><strong>Run your setup on a GPU generation it never ran on.</strong> Renest doesn't
  swap your PyTorch version. Get it working on the new card, then pack it there. See
  <a href="/answers/errors-after-gpu-change">Errors after changing GPU</a>.</li>
  <li><strong>Repair a broken rented machine</strong>, such as a GPU that doesn't answer or
  a driver that's too old. The fix is another machine, and the same restore command works
  there.</li>
  <li><strong>Fix a workflow file someone sent you.</strong> See
  <a href="/answers/red-nodes-empty-models">Red nodes and empty model lists</a>.</li>
  <li><strong>Make downloads faster.</strong> The models still have to come down onto the
  new machine. What a restore saves is the reinstalling and the trial and error.</li>
  <li><strong>Replace a network volume you're happy with.</strong> If you always come back
  to the same region of the same provider and your setup survives restarts, keep using
  it.</li>
</ul>

<p>If none of these is your situation, start with
<a href="/answers/move-comfyui-to-another-machine">Move ComfyUI to another machine</a> or
<a href="/answers/when-is-a-move-finished">When is a move actually finished?</a></p>