How to use a Flux LoRA (and a Flux 2 LoRA) on Graydient: the beginner guide

A Flux LoRA is a small add-on file that teaches the Flux image model one thing: a face, an art style, a camera look, a piece of clothing, a fix for hands. Thousands exist for Flux.1 and for Flux 2, and most of them are free. This guide shows a complete beginner how to get one into Graydient and use it, with no files to download and no graphics card. Everything runs on our GPUs for one flat monthly price.

You will do three things: pick a LoRA, add it to your Graydient library (one click from ModelTrends, a Civitai link, or a Hugging Face file), and call it in a prompt. Each step is written out below. If you already know how to import, jump to the table that shows how every Graydient app calls a LoRA.

What you need

• A Graydient plan. Images is enough for pictures; Video Plus or Video Max add video, upscaling and editing. Compare plans and pricing, or log in to Graydient if you already have one.
• A browser, or Telegram or Discord on your phone. A Hugging Face account only when the LoRA you want is stored there.
• Ten minutes. A small LoRA is usually ready to render a few minutes after you add it.

Flux LoRA basics: the one rule that matters

A LoRA never renders alone. It rides on a base model, and it only works on the base family it was trained for. On Graydient the base is a ShortCode you add to the prompt, so the rule is simple:

• A Flux.1 LoRA (made for Flux.1 dev, schnell or the Flux Krea dev base) renders with <flux-dev> or another Flux.1 checkpoint from your library.
• A Flux 2 LoRA (made for Flux 2 dev or Flux 2 Klein) renders with <flux2> or a Flux 2 based checkpoint. Flux.1 and Flux 2 LoRAs are not interchangeable.
• Krea 2 is its own model, not Flux. Flux LoRAs do not apply to it. The ModelTrends page of each LoRA names the family, so you never have to guess.
• Flux reads natural sentences and ignores negative prompts. Describe the scene, add the trigger word from the LoRA page, and leave the old Stable Diffusion keyword lists behind.

Way 1: one click from ModelTrends (easiest)

ModelTrends is the catalog of models Graydient hosts. Every Flux and Flux 2 LoRA with a ShortCode on its page can be added to your library without touching a file.

1. Open the Flux or Flux 2 family on ModelTrends and filter by LoRA. Sort by trending or search a name. Checkpoints and LoRAs are marked, so pick a LoRA card.
2. Open the LoRA page. Write down three things: the trigger word (if any), the recommended weight, and the ShortCode, written like <color-grading-flux>.
3. Press the install control on that page while you are logged in to Graydient. Confirm if asked. The full walkthrough is how to install models from ModelTrends on Graydient.
4. Wait until the LoRA shows in your Graydient library. Small LoRAs are often ready in a few minutes.
5. Copy the ShortCode. You will paste it in a prompt in the next section.

Way 2: paste a Civitai link

Use this when the LoRA is on Civitai but not on ModelTrends yet. You stay on Civitai's free side: nothing here spends Buzz.

1. On Civitai, open the LoRA and check the base model line: it must say Flux.1 D, Flux.1 S, Flux Krea or Flux 2. Copy the page URL from your browser.
2. log in to Graydient, open Model Import from your dashboard library, and paste the URL. If the page lists several versions, pick the one you want before you confirm.
3. Submit and wait for the LoRA to appear in your library. Copy the trigger word and weight from the Civitai page while it loads.
4. The library shows the ShortCode Graydient gave the LoRA. Use it exactly as written. The full guide to this path is how to import Civitai and Hugging Face models on Graydient.

Way 3: a Hugging Face file

Many Flux LoRA trainers publish on Hugging Face, and some Civitai files are mirrored there. Model Import takes a direct link to the .safetensors file.

1. Open the repository, click Files, find the .safetensors LoRA, and copy its download link (the resolve address). Accept the license on Hugging Face first if the repository is gated.
2. Paste the link into Model Import in your Graydient dashboard. If a Hugging Face sign-in window opens, log in there; Graydient only fetches files you are allowed to download.
3. Wait for the LoRA in your library and copy its ShortCode.

Prefer Telegram? You can also register your own Hugging Face file from the chat: install a Civitai model that is not on ModelTrends (PirateDiffusion guide).

Render with the LoRA in every Graydient app

Once a LoRA is in your library, the same tag works everywhere. The tag has three parts: the word lora, the ShortCode, and a weight between 0 and 1, like <lora:shortcode:0.8>. Then name the base. Here is how each app takes it:

A complete Flux.1 example for Telegram or Discord, with a colour-grading LoRA at 0.8 on the Flux dev base: /render cinematic style, a woman reading by a rainy window, 35mm photo <lora:color-grading-flux:0.8> <flux-dev> /steps:24 /guidance:3.5 /size:1024x1024

The same idea on Flux 2, with a Flux 2 LoRA and the Flux 2 base: /render a man waiting for a bus at dawn, 90s indie movie look <lora:indie-movie-90s-flux2:0.7> <flux2> /size:1024x1024

Weights, trigger words and stacking

The number after the ShortCode is the weight. It is the one dial that beginners miss, and it fixes most bad renders.

• Start at the weight the LoRA page recommends; 0.8 is a safe default when the page says nothing. Flux LoRAs tend to be strong, so 1.0 is often too much.
• Faces look waxy, colours blow out, or the picture turns to noise: lower the weight in steps of 0.1 until it comes back into focus.
• The effect is too weak: raise the weight a little, and make sure the trigger word from the LoRA page is in the prompt. Many style LoRAs do little without it.
• Stack two or three LoRAs by adding more tags, each with its own weight, before the base: <lora:first-lora:0.6> <lora:second-lora:0.6> <flux-dev>. Keep each at 0.7 or lower so they do not fight.
• Flux.1 dev likes about 20 to 28 steps and a guidance near 3.5. Flux 2 Klein based models are distilled and finish in a few steps at guidance 1. The model page lists what its author recommends.

Flux 2 LoRAs: what is different

Flux 2 is a new architecture, so its LoRAs are a separate set. The workflow on Graydient is the same, with four things to know:

• Use the Flux 2 base, <flux2>, or a Flux 2 based checkpoint. A Flux.1 LoRA on a Flux 2 base does nothing useful, and the reverse is the same.
• Flux 2 Klein is the fast, distilled branch. Klein LoRAs and Klein based checkpoints render in a few steps; do not copy Flux.1 step counts into them.
• Flux 2 understands longer, more literal prompts and reads text in the picture well. Describe materials, light and lens as a photographer would.
• On ModelTrends the Flux 2 family has its own page and its own LoRA list, so you can tell at a glance which set a LoRA belongs to.

If something looks wrong

• The LoRA seems ignored: check the family. A Flux.1 LoRA with <flux2> (or the other way round) is the usual cause. Fix the base ShortCode.
• The picture is fried or grey: the weight is too high, or two strong LoRAs are stacked at 1.0. Lower them.
• Nothing renders and the app reports an unknown model: the LoRA is still loading, or the ShortCode has a typo. Open your library and copy the code again.
• The style is right but the subject is wrong: Flux follows the sentence. Put the subject first, the style words and trigger after it, and drop keyword lists.
• You pasted a Civitai prompt with brackets: Graydient uses square brackets for negatives and Flux ignores negatives anyway. Delete them.

That is the whole method: add the LoRA once, then write prompts with its tag and the right base. Because the plan is flat, you can try ten weights and ten seeds for one picture without watching a balance. When you are ready, log in to Graydient or compare plans and pricing. Every plan includes a 7-day money-back guarantee.

Questions

See plans and pricing